BEGIN:VCALENDAR VERSION:2.0 PRODID:-//HasGeek//NONSGML Funnel//EN DESCRIPTION:AI at the heart of industry & innovation X-WR-CALDESC:AI at the heart of industry & innovation NAME:The Fifth Elephant - Pune Edition X-WR-CALNAME:The Fifth Elephant - Pune Edition REFRESH-INTERVAL;VALUE=DURATION:PT12H SUMMARY:The Fifth Elephant - Pune Edition TIMEZONE-ID:Asia/Kolkata X-PUBLISHED-TTL:PT12H X-WR-TIMEZONE:Asia/Kolkata BEGIN:VEVENT SUMMARY:Workshop: Build & Optimise AI Agents That Survive Production DTSTART:20260227T043000Z DTEND:20260227T073000Z DTSTAMP:20260421T120126Z UID:session/NocANVGc1vbMTaBccf5Gr7@hasgeek.com SEQUENCE:8 CATEGORIES:Track 1 AI in Software Development Life Cycle (SDLC),Hands-on w orkshop (1.5 hrs to 3 hrs) CREATED:20260203T141532Z DESCRIPTION:### Description\n\nIt’s easy to ship a magical agent demo. I t’s much harder to ship an agent that works for real users: noisy inputs \, partial context\, flaky tools\, ambiguous goals\, and “tiny prompt ch anges” that break everything.\n\nIn this hands-on workshop\, we’ll bui ld a small but realistic agent in **Python + DSPy**\, then turn it into so mething you can actually run in production: **structured I/O**\, **tool co ntracts**\, **tracing**\, **evals**\, and **automatic optimisation**.\n\nY ou’ll leave with a concrete engineering workflow (an “agent improvemen t loop”) that you can take back to your team:\n\n**instrument → collec t failures → convert to evals → optimise → ship via CI**.\n\n> The t echniques are framework-agnostic\; we’ll use DSPy because it makes optim isation and modularity explicit in code.\n> \n\n### Key Takeaways\n\n- A p ractical mental model of agents: **goal → plan/act loop → tool calls → ground-truth checks → stop conditions**.\n- How to build agents as * *maintainable software** (signatures/modules) instead of brittle prompt bl obs.\n- How to add **observability + evals** so you can debug “why it fa iled” and measure progress.\n- How to use **DSPy optimisers** (few-shot + program/prompt optimisation) to improve quality *systematically*.\n- A r epeatable **CI workflow** to keep agents improving safely as users and req uirements change.\n\n### Target Audience\n\n- **Level:** Intermediate\n- * *Prerequisites:**\n - Comfortable with Python (APIs\, functions\, virtu alenv/uv)\n - Basic familiarity with LLMs (prompts\, tool calling conce pts)\n - Laptop + internet access\n- **Best suited for:**\n - Softwa re engineers / platform engineers building LLM features\n - SDETs / QA engineers working on evals and reliability\n - Engineering managers and tech leads who need a production-ready approach\n\n### Workshop outline ( 3 hours)\n\n1. **Anatomy of a production agent (15 min)**\n \n Agent loop\, tool contracts\, ground-truth checks\, stop conditions\, failure m odes.\n \n2. **Build the agent in DSPy (60 min)**\n \n Signatures + modules\, tool wiring\, structured outputs\, error handling.\n \n3. **Observability & evals (45 min)**\n \n Tracing\, failure buckets\, creating an eval set from real-ish cases\, measuring baseline.\n \n4. * *Optimisation (45 min)**\n \n Few-shot baselines → DSPy optimiser run → compare metrics + inspect deltas.\n \n5. **Shipping the improve ment loop (15 min)**\n \n Minimal CI pattern: run evals on PR\, regr essions gate merges\, version prompts/programs.\n \n\n### Setup\n\nThe workshop skeleton and requirements can be found in this repo: https://gith ub.com/unravel-team/real-agents-workshop\n\n### Speaker bio (short)\n\n**K iran Kulkarni** is the founder of **Unravel.tech**\, where he helps teams build **production-grade AI systems**—agentic workflows\, evaluation pip elines\, and reliability/observability practices. He’s been a founding e ngineer and engineering leader across data + AI systems and loves turning “cool demos” into software that survives real users.\n\n**Utkarsh Digh e** is a senior engineer at **Unravel.tech**\, where he designs and builds pragmatic solutions using Agentic AI to tackle complex problems across do mains. He takes an engineering-first approach\, focusing on reliability\, robustness\, and scalability—ensuring systems don’t just work in theor y\, but hold up under real-world usage. GEO:18.571009876468388;73.77414356162001 LAST-MODIFIED:20260224T062332Z LOCATION:Ursa Major - Nutanix Technologies India Pvt Ltd\nPune\nIN ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com URL:https://hasgeek.com/fifthelephant/2026-pune/schedule/workshop-build-op timise-ai-agents-that-survive-production-NocANVGc1vbMTaBccf5Gr7 BEGIN:VALARM ACTION:display DESCRIPTION:Workshop: Build & Optimise AI Agents That Survive Production i n Ursa Major in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT BEGIN:VEVENT SUMMARY:Fine-Tuning SLM models with LoRA: Building Specialized On-Device S tory Generators DTSTART:20260227T050000Z DTEND:20260227T073000Z DTSTAMP:20260421T120126Z UID:session/5rscwBW3Tv2twKy4a8Knot@hasgeek.com SEQUENCE:12 CATEGORIES:Hands-on workshop (1.5 hrs to 3 hrs) CREATED:20260203T141817Z DESCRIPTION:# Fine-Tuning SLMs with LoRA: Building Specialized On-Device S tory Generators\n## Description\nAs generative AI moves closer to the edge \, developers are looking for ways to build creative\, high-quality applic ations that run privately\, efficiently\, and without dependence on cloud APIs. This session explores how to shift from server-side story generation using large language models (LLMs) to highly optimized on-device workflow s powered by small language models (SLMs). Attendees will learn the end-to -end process of generating content in the cloud\, reproducing it locally\, and progressively improving output quality using prompt tuning and LoRA-b ased fine-tuning.\n\nThrough a series of practical demonstrations\, we wil l walk through three stages of on-device model refinement: baseline infere nce\, prompt-tuned enhancement\, and LoRA-based adapter tuning for persona lization. Participants will compare outputs from each stage\, understand t he trade-offs in quality vs. performance\, and learn lightweight evaluatio n methods for generative storytelling. By the end\, they will know how to build efficient\, privacy-preserving\, specialized story generators that c an run directly on mobile or embedded devices.\n\n## Key Takeaways\n- Lear n how to migrate generative workflows from cloud LLMs to optimized\, on-de vice SLMs.\n- Understand and apply prompt tuning and LoRA adapter tuning t o personalize model behavior.\n- Gain practical methods to evaluate improv ements in on-device generative quality.\n\n## Target Audience\n- Level: In termediate\n- Prerequisites:\n - Basic Python knowledge\; familiarity w ith foundational LLM concepts\n - Laptop and internet connectivity!\n- Best suited for:\n - AI/ML developers building edge or offline applicatio ns\n - Mobile developers exploring on-device inference\n - Researchers w orking with small\, efficient model architectures\n - Engineers evaluatin g personalization strategies for constrained devices\n\n---\n\nMayur is a seasoned engineer specializing in AI\, data\, and backend systems\, with e xtensive experience building scalable\, high-performance platforms at orga nizations such as JioHotstar\, Intuit\, Walmart\, and SAP. He frequently d elivers webinars and technical sessions on AI engineering and distributed systems\, and actively shares insights with the developer community. Conne ct with him at: https://www.linkedin.com/in/mayurmadnani/ GEO:18.571009876468388;73.77414356162001 LAST-MODIFIED:20260224T062342Z LOCATION:Ursa Minor - Nutanix Technologies India Pvt Ltd\nPune\nIN ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com URL:https://hasgeek.com/fifthelephant/2026-pune/schedule/fine-tuning-slm-m odels-with-lora-building-specialized-on-device-story-generators-5rscwBW3Tv 2twKy4a8Knot BEGIN:VALARM ACTION:display DESCRIPTION:Fine-Tuning SLM models with LoRA: Building Specialized On-Devi ce Story Generators in Ursa Minor in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT BEGIN:VEVENT SUMMARY:Lunch break DTSTART:20260227T073000Z DTEND:20260227T083000Z DTSTAMP:20260421T120126Z UID:session/Lut8H2GDvpqyZMP9QNaJ2X@hasgeek.com SEQUENCE:3 CREATED:20260203T141901Z LAST-MODIFIED:20260203T141912Z LOCATION:Pune ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com BEGIN:VALARM ACTION:display DESCRIPTION:Lunch break in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT BEGIN:VEVENT SUMMARY:Accelerating Agentic AI Adoption with Nutanix Enterprise AI - work shop DTSTART:20260227T083000Z DTEND:20260227T110000Z DTSTAMP:20260421T120126Z UID:session/D5BZN2VQ8Lh4DATdBtNNzN@hasgeek.com SEQUENCE:8 CATEGORIES:Track 1 AI in Software Development Life Cycle (SDLC),Hands-on w orkshop (1.5 hrs to 3 hrs) CREATED:20260203T141924Z DESCRIPTION:AI Agents can autonomously think\, plan and execute with minim al human supervision. AI Agents have shown promising results in improving productivity which is why enterprises are eager to adopt AI agents. AI Age nts are designed to handle complex tasks using the underlying model’s re asoning capabilities which are context sensitive and probabilistic. This m eans sometimes\, they can exhibit non-deterministic behaviours.\n\nEnterpr ises can adopt AI Agents only after addressing the Security\, Compliance a nd Operational concerns with them. They would need full visibility and con trol over what AI Agents do. There have been cases where Agents have leake d sensitive information or misconfigured environments. Ensuring AI Agents comply with the organization’s IT and Data Access policies is necessary. Building an AI agent that works in a demo is easy. Building one that ente rprises can trust in production is a different challenge altogether.\n\nIn this workshop\, we'll start with the fundamentals of AI agents\, LLMs as the reasoning engine and tools as the execution layer. We will then focus on what separates an enterprise-ready deployment from a proof-of-concept: authentication\, role-based access control (RBAC)\, rate limiting\, observ ability and audit trails. And how these will help enterprises overcome the challenges with AI Agent Adoption. You'll learn how to leverage Nutanix E nterprise AI (NAI) to deploy production-grade AI agents with minimal opera tional overhead. We'll walk through the platform's streamlined workflows f or Model Deployment and Access Control\, Secure Inference API Access\, MCP (Model Context Protocol) Server Integration\, and Regulated MCP Tool Perm issions.\n\nBy the end of this workshop\, you'll have a clear blueprint fo r deploying AI agents that meet enterprise security and compliance require ments\, maintaining developer velocity.\n\n## Key takeaways\n* Understand the core principles of AI agents and how they function.\n* Explore the cur rent challenges enterprises face with adoption of AI Agents and learn how Nutanix Enterprise AI can help address them.\n* Gain practical knowledge o n building and deploying AI Agents leveraging Nutanix Enterprise AI.\n\n** Level:** Intermediate\n\n**Prerequisites:**\n* Basic understanding of LLMs and how Inference works\n* Basic understanding of MCP\n* Hardware/Softwar e requirements:\nMac or Windows computer\nLatest version of Python and uv installed\n\n## Target audience\n* Software engineers building AI applicat ions for enterprise use\n* IT Admins managing AI Infrastructure and MCP Se rvers\n* Security and compliance teams assessing AI deployment risks\n* En gineering managers and architects designing AI Agents and AI Platforms\n\n ## Instructors\n**Aishwarya Raimule** is an ML Systems Engineer at Nutanix \, India where she is currently working on Observability for Nutanix AI In ference Platform for LLMs and MCP Integration for Agentic workflows.\n\n** Hritik Raj** is an ML Systems Engineer at Nutanix since an year and half. He has been working on AI observability and LLM benchmarking initiatives. He is interested in Agentic orchestrators\, Finetuning and LLM evaluations .\n\n GEO:18.571009876468388;73.77414356162001 LAST-MODIFIED:20260224T062352Z LOCATION:Ursa Major and Ursa Minor - combined - Nutanix Technologies India Pvt Ltd\nPune\nIN ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com URL:https://hasgeek.com/fifthelephant/2026-pune/schedule/accelerating-agen tic-ai-adoption-with-nutanix-enterprise-ai-workshop-D5BZN2VQ8Lh4DATdBtNNzN BEGIN:VALARM ACTION:display DESCRIPTION:Accelerating Agentic AI Adoption with Nutanix Enterprise AI - workshop in Ursa Major and Ursa Minor - combined in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT BEGIN:VEVENT SUMMARY:Check-in DTSTART:20260228T033000Z DTEND:20260228T041500Z DTSTAMP:20260421T120126Z UID:session/8XS63b898THJcUkrgZwwHn@hasgeek.com SEQUENCE:10 CREATED:20260203T120837Z LAST-MODIFIED:20260208T063853Z LOCATION:Pune ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com BEGIN:VALARM ACTION:display DESCRIPTION:Check-in in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT BEGIN:VEVENT SUMMARY:Conference introductions DTSTART:20260228T041500Z DTEND:20260228T042500Z DTSTAMP:20260421T120126Z UID:session/46f2KipntJG2AmXMj3sgNb@hasgeek.com SEQUENCE:4 CREATED:20260203T120909Z GEO:18.571009876468388;73.77414356162001 LAST-MODIFIED:20260208T064146Z LOCATION:Ursa Major and Ursa Minor - combined - Nutanix Technologies India Pvt Ltd\nPune\nIN ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com BEGIN:VALARM ACTION:display DESCRIPTION:Conference introductions in Ursa Major and Ursa Minor - combin ed in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT BEGIN:VEVENT SUMMARY:Agentic intelligence fabric for smart buildings DTSTART:20260228T043000Z DTEND:20260228T050500Z DTSTAMP:20260421T120126Z UID:session/XNpAbJmSFCVmDivoCbbVxn@hasgeek.com SEQUENCE:21 CATEGORIES:Track 2: AI in Manufacturing & Digital Transformation,30 mins t alk CREATED:20260203T121028Z DESCRIPTION:**Session Overview**\n\nSmart buildings today generate massive volumes of energy\, occupancy\, environmental\, and asset data—but orga nizations still struggle to turn these signals into reliable\, automated d ecisions. Traditional analytics or rules-based workflows cannot keep up wi th the complexity\, leading to inconsistent operations\, missed insights\, and systems that are difficult to scale or trust in mission-critical envi ronments.\n\nThis session introduces the Agentic Intelligence Fabric\, a n ew architectural approach that brings trustworthy and coordinated AI intel ligence to modern building systems. By combining multi-agent collaboration \, structured memory\, and built-in governance mechanisms\, the framework enables AI agents to reason across building subsystems\, detect issues ear lier\, and support high-quality decisions without constant human oversight . This creates a foundation for stable and consistent automation across en ergy\, occupancy\, environmental\, and asset domains.\n\nAttendees will le arn how real-time telemetry moves through this architecture\, how agents c oordinate during complex workflows\, and how the system enables safer\, mo re reliable\, and more efficient building operations at scale. The approac h is designed to align with real-world enterprise requirements and demonst rates a practical path toward next-generation building intelligence.\n\n** Key Takeaways**\n\nHow multi-agent AI systems can unify intelligence acros s energy\, occupancy\, environmental\, and asset operations within a singl e automation layer.\n\nHow trust layers and memory structures reduce incon sistencies and support reliable\, repeatable AI-driven workflows.\n\nHow r eal-time building data combined with coordinated AI agents enables proacti ve optimization\, anomaly detection\, and mission-critical decision suppor t.\n\n**Target Audience**\n\n- AI and LLM practitioners interested in agen t-based architectures for real-world deployments.\n\n- Smart-building and IoT teams seeking more reliable and automated decision systems.\n\n- Facil ity\, energy\, and operations leaders looking for scalable intelligence ac ross complex building portfolios.\n\n**Speaker Bio**\n\nHimanshu Agrawal i s an AI Data Scientist at Johnson Controls\, where he develops advanced AI -driven solutions for smart-building ecosystems. His work spans multi-agen t architectures\, predictive modeling\, and intelligent automation. He bri ngs hands-on experience across large language models\, applied machine lea rning\, and distributed AI systems. Himanshu holds a Master’s degree in Data Science from IIIT Bangalore and a Bachelor’s degree in Computer Sci ence. GEO:18.571009876468388;73.77414356162001 LAST-MODIFIED:20260222T130608Z LOCATION:Ursa Major and Ursa Minor - combined - Nutanix Technologies India Pvt Ltd\nPune\nIN ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com URL:https://hasgeek.com/fifthelephant/2026-pune/schedule/agentic-intellige nce-fabric-for-smart-buildings-XNpAbJmSFCVmDivoCbbVxn BEGIN:VALARM ACTION:display DESCRIPTION:Agentic intelligence fabric for smart buildings in Ursa Major and Ursa Minor - combined in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT BEGIN:VEVENT SUMMARY:Interrogating your twin: causal reasoning in manufacturing systems DTSTART:20260228T051000Z DTEND:20260228T054500Z DTSTAMP:20260421T120126Z UID:session/WsyBSFbVAoprBKx3YJWPCE@hasgeek.com SEQUENCE:10 CATEGORIES:Track 2: AI in Manufacturing & Digital Transformation,30 mins t alk CREATED:20260203T120953Z DESCRIPTION:\n**Describe your session in 2 paragraphs**\n\nDigital twins\, predictive maintenance models\, and AI-driven quality control all promise to tell manufacturers what to do. But there's a critical gap - most ML mo dels only reveal correlations\, not causation. When your predictive mainte nance system flags high vibration as a failure predictor\, does reducing v ibration actually prevent failures\, or is vibration merely a symptom of t he real root cause like bearing wear or misalignment? When your quality mo del correlates temperature with defect rates\, will adjusting temperature fix the problem\, or is temperature confounded by shift changes and raw ma terial batches? This distinction between "what predicts" and "what to inte rvene on" is the difference between optimizing your factory and chasing ex pensive red herrings.\n\nThis talk introduces Pearl's ladder of causation as a practical framework for interrogating your Industrie 4.0 analytics— whether that's a full digital twin or sensor data feeding a maintenance da shboard. Using a predictive maintenance scenario\, I'll demonstrate how ca usal inference tools—DAGs\, the do-calculus\, and adjustment sets—dist inguish spurious correlations from actionable interventions. I'll show why LLMs and standard ML are "causal parrots" that confuse P(Y|X) with P(Y|do (X))\, and how to augment them with proper causal machinery. The result: e xplicit\, auditable assumptions that let operations\, engineering\, and da ta science teams align on why an intervention should work before committin g resources.\n\n **Mention 1-2 takeaways from your session**\n\n1. A clea r framework (Pearl's ladder) for recognizing when manufacturing decisions require causal inference rather than correlation-based ML—especially for root cause analysis\, intervention planning\, and counterfactual "what-if " scenarios\n2. A practical workflow to encode domain knowledge as testabl e causal graphs (DAGs) and derive statistically valid adjustment strategie s for predictive maintenance and quality control problems\n\n **Which aud iences is your session going to be beneficial for?**\n\n Data scientists\ , analytics managers\, and manufacturing/operations engineers building pre dictive maintenance systems\, quality control models\, or digital twins wh o want to move beyond correlation-based dashboards to causal\, interventio nal decision-making. Also valuable for managers navigating disagreements b etween AI recommendations and shop-floor intuition.\n\n**Bio:** I am a lap sed physicist with several years of experience in various Data Science and BI contexts\, mostly in Germany. Now I am the founder of Romulan AI - bui lding the causal layer for AI based digital transformation. GEO:18.571009876468388;73.77414356162001 LAST-MODIFIED:20260222T130846Z LOCATION:Ursa Major and Ursa Minor - combined - Nutanix Technologies India Pvt Ltd\nPune\nIN ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com URL:https://hasgeek.com/fifthelephant/2026-pune/schedule/interrogating-you r-twin-causal-reasoning-in-manufacturing-systems-WsyBSFbVAoprBKx3YJWPCE BEGIN:VALARM ACTION:display DESCRIPTION:Interrogating your twin: causal reasoning in manufacturing sys tems in Ursa Major and Ursa Minor - combined in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT BEGIN:VEVENT SUMMARY:From scratch to production in three months: the AI-first delivery model DTSTART:20260228T055000Z DTEND:20260228T062000Z DTSTAMP:20260421T120126Z UID:session/6N8XPLwNtrJV8GksqE9Zf5@hasgeek.com SEQUENCE:9 CATEGORIES:Track 1 AI in Software Development Life Cycle (SDLC),30 mins ta lk CREATED:20260203T121543Z DESCRIPTION:# Session Description\n\nMost teams bolt AI onto existing proc esses: same Jira workflows\, same sprint rituals\, same onboarding docs\, just with Copilot autocompleting some code. But what if you could start fr esh? What if you had a greenfield project and could design your entire del ivery process knowing what AI can do today?\n\nWe got that chance. Equal E xperts and Travelopia partnered to rebuild a legacy lead scoring system fr om scratch. No legacy code to migrate\, no existing architecture to preser ve\, just a business case and a blank repo. We used this clean slate to as k a dangerous question: what would an SDLC look like if we designed it aro und AI from day one? The answer looked nothing like what we'd been taught. Engineers and Product Owners paired together to write PRD documents in ma rkdown\, committed directly to the repo\, creating a shared context that b oth humans and LLMs could consume. We generated stakeholder reports direct ly from commit histories\, and structured documentation so well that new d evelopers achieved fully self-service onboarding without formal KT session s. Three months later\, we shipped to production. Seven months on\, the sy stem handles 5\,000+ leads monthly.\n\nThis talk isn't about adding AI to your workflow. It's about what becomes possible when you design the workfl ow around AI. I'll share the specific patterns we used\, the guardrails th at kept AI from over-engineering (human-owned architecture\, open-source t ools like Context7 for up-to-date library documentation)\, and a practical framework for teams ready to rethink\, not just optimise\, how they deliv er software.\n\n---\n\n## Key Takeaways\n\n1. **A blueprint for AI-native delivery when starting fresh**: Learn how to design a ceremony-light\, Git -native SDLC from scratch\, where engineers and Product Owners collaborate on markdown PRDs in the repo\, commit histories drive stakeholder reporti ng\, and documentation doubles as AI context and self-service onboarding m aterial.\n\n2. **Guardrails for letting AI do the heavy lifting without lo sing control**: Understand how to draw clear boundaries between human deci sions (architecture\, security\, design) and AI execution (implementation\ , tests\, documentation)\, with open-source tooling patterns that keep you r AI current and your codebase maintainable.\n\n---\n\n## Target Audience\ n\nThis session is for engineering leads\, architects\, and senior develop ers who are about to start a new project or rewrite and are wondering: sho uld we do this differently? It's also for delivery managers frustrated wit h process overhead who suspect there's a better way but haven't seen it pr oven. You've experimented with AI coding tools but haven't had the chance to design a delivery process around them. You'll leave with a concrete mod el for what's possible when you start from zero.\n\n---\n\n## Speaker Bio\ n\nPrasanth Jayaprakash is a Principal Consultant at Equal Experts with ne arly 17 years of hands-on experience building production systems across ba ckend (Java\, Kotlin\, Groovy) and frontend (React\, Vue) technologies.\n\ nHe recently led an experimental AI-native delivery engagement with Travel opia\, testing whether AI could fundamentally change how teams coordinate work\, not just how fast they write code. The result: a lead-scoring platf orm built in three months\, running in production for seven months without major issues.\n\nHis current work focuses on two areas: transforming soft ware delivery practices using LLM-powered workflows\, and building agentic AI systems using RAG and tool integration. This talk shares the operation al framework\, specific guardrails\, and honest assessment of when ceremon y-light delivery works and when it doesn't.\n\n## Reference\nhttps://www.e qualexperts.com/case-study/engineering-ai-into-software-delivery-how-trave lopia-launched-software-to-production\n GEO:18.571009876468388;73.77414356162001 LAST-MODIFIED:20260222T130910Z LOCATION:Ursa Major and Ursa Minor - combined - Nutanix Technologies India Pvt Ltd\nPune\nIN ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com URL:https://hasgeek.com/fifthelephant/2026-pune/schedule/from-scratch-to-p roduction-in-3-months-the-ai-first-delivery-model-6N8XPLwNtrJV8GksqE9Zf5 BEGIN:VALARM ACTION:display DESCRIPTION:From scratch to production in three months: the AI-first deliv ery model in Ursa Major and Ursa Minor - combined in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT BEGIN:VEVENT SUMMARY:Birds of Feather session (BOF) (Sponsored): LLM inference optimiza tions: a deep dive into modern techniques DTSTART:20260228T060000Z DTEND:20260228T070000Z DTSTAMP:20260421T120126Z UID:session/VPZv2S5hJ6QGxGWMnk4aRn@hasgeek.com SEQUENCE:35 CATEGORIES:Track 1 AI in Software Development Life Cycle (SDLC),Birds of F eather (BOF) session CREATED:20260203T123758Z DESCRIPTION:## Problem statement\nThe core problem discussed is the "Memor y Wall" in LLM inference—where GPU computational power has scaled dramat ically (~50\,000x+ in the last decade)\, but memory bandwidth has lagged ( only 100x growth)\, making inference memory-bound rather than compute-boun d. This leads to idle GPU cores\, high latency\, and inefficient resource utilization\, especially for long-context models and batch processing.\n\n Under this topic\, we intend to cover a few popular techniques on improvin g memory usage efficiency such as the following to unlock the LLM potentia ls for:\n* Flash Attention\n* Virtual memory-inspired techniques to elimin ate fragmentation\, Paged Attention and Prefix Caching\n* Use of KV Caches and KV Cache Compression\n* Continuous Batching and Speculative Decoding to Alleviate bandwidth bottlenecks and improve the compute-to-memory movem ent ratio\n\n## Key takeaways\n1. Attendees will gain a clear understandin g of why LLM inference is memory-bound and how use of kv cache and techniq ues like Flash Attention and Paged Attention can achieve 2-4x speedups and higher GPU utilization\, enabling longer contexts and larger batches with out hardware upgrades.\n2. Participants will learn actionable strategies f or KV cache management and speculative decoding\, leading to faster token generation (~2x - 3x)while maintaining equivalence to standard methods\, d irectly applicable to real-world serving systems like vLLM.\n\n## Audience s for this session\nThis discussion will benefit:\n* Machine learning engi neers and AI developers involved in deploying and scaling LLMs in producti on environments\, who need practical techniques to reduce latency and cost s.\n* Researchers and data scientists focused on transformer architectures \, seeking insights into memory bottlenecks and optimization trade-offs.\n * Product managers and tech leads in AI-driven companies (e.g.\, chatbots\ , recommendation systems)\, who can apply these efficiencies to improve th roughput and user experience.\n\n## About the facilitator\nKundan Kumar is a final year Computer Science student at IIT Kanpur. He has worked on KV caching systems at Nutanix as a visiting researcher. His interests lie at the intersection of systems optimization and AI infrastructure. GEO:18.571009876468388;73.77414356162001 LAST-MODIFIED:20260222T131238Z LOCATION:Alpha Century room for BOFs - Nutanix Technologies India Pvt Ltd\ nPune\nIN ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com URL:https://hasgeek.com/fifthelephant/2026-pune/schedule/llm-inference-opt imizations-a-deep-dive-into-modern-techniques-VPZv2S5hJ6QGxGWMnk4aRn BEGIN:VALARM ACTION:display DESCRIPTION:Birds of Feather session (BOF) (Sponsored): LLM inference opti mizations: a deep dive into modern techniques in Alpha Century room for BO Fs in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT BEGIN:VEVENT SUMMARY:Morning break DTSTART:20260228T062000Z DTEND:20260228T064000Z DTSTAMP:20260421T120126Z UID:session/XfyyoJBSsLBtnCMV4mABnV@hasgeek.com SEQUENCE:5 CREATED:20260203T121555Z GEO:18.571009876468388;73.77414356162001 LAST-MODIFIED:20260208T064333Z LOCATION:Ursa Major and Ursa Minor - combined - Nutanix Technologies India Pvt Ltd\nPune\nIN ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com BEGIN:VALARM ACTION:display DESCRIPTION:Morning break in Ursa Major and Ursa Minor - combined in 5 min utes TRIGGER:-PT5M END:VALARM END:VEVENT BEGIN:VEVENT SUMMARY:Sponsored talk: Agentic triage engine DTSTART:20260228T064000Z DTEND:20260228T071000Z DTSTAMP:20260421T120126Z UID:session/AQnqnC8W4ZgwfXywonRmHj@hasgeek.com SEQUENCE:19 CATEGORIES:Track 1 AI in Software Development Life Cycle (SDLC),30 mins ta lk CREATED:20260203T121502Z DESCRIPTION:### Challenge\nComplex software products span a wide range of capabilities - hardware and software - hypervisors\, containers\, storage\ , servers\, networking\, systems\, control plane\, management plane\, secu rity - for on-prem and public cloud systems.\n\nSoftware Development lifec ycle includes the qualification of product features\, as they are develope d\, using tests in isolated and integrated environments - to test individu al functionality\, the overall system\, in happy path and negative scenari os\, as well as cross-functional workflows involving multiple layers of so ftware.\n\nSuch tests are executed frequently (daily\, weekly etc) to dete ct breakages early in the software development lifecycle. Test failures ne ed to be triaged efficiently\, for maximum benefit.\n\nAgentic Triage Engi ne uses Agentic and AI tools & techniques to efficiently triage test failu res\, by integrating into the Quality Assurance (QA) infrastructure\, soli cit and incorporate QA engineers’ review and feedback to reduce overall time to triage failures.\n\n### Key Takeaways\n- Agentic and AI approaches for Efficiency in Triage.\n- Determine duplicate vs fresh failures quickl y.\n- Integrate Domain Expert (QA Engineer) feedback seamlessly.\n\n### Ta rget Audience\n- Software Engineers and AI Developers directly involved in software development lifecycle.\n- Verification and Validation Engineers responsible for Qualification of Software products.\n- Engineering Manager s and Tech Leads responsible for bringing efficiency in software developme nt lifecycle\n\n### Speakers Bio\n**Pingjin Wen**: Pingjin Wen is a Senior Staff Engineer at Nutanix with extensive experience in testing large-scal e distributed systems and building high-impact test automation platforms. His current work centers on bringing AI into the QA lifecycle\, enabling d ata-driven quality signals\, automated insights from test failures\, and m ore efficient end-to-end SDLC workflows.\n\n**Geetha Srikantan** Geetha Sr ikantan contributed to early AI/ML projects for the USPS\, IRS and other f ederal agencies\, and has worked on several products spanning Streaming Me dia\, Hyper-Converged Infrastructure\, Data Protection. More recently\, sh e has been involved in Root Cause Analysis projects at Nutanix.\n\n\n**Tin gting Li**: Tingting Li is a Lead Engineer in the System Test Core Data Pa th team. She is responsible for ensuring the quality of the Cloud Native A OS product. One of the team’s founding engineers initiated the Triage Ge nie AI project\, leveraging AI technologies such as Large Language Models (LLMs)\, Retrieval-Augmented Generation (RAG)\, and AI-driven workflows to automate the QA triage and test process. A patent is currently pending fo r the LogRAG project.\n GEO:18.571009876468388;73.77414356162001 LAST-MODIFIED:20260228T062445Z LOCATION:Ursa Major and Ursa Minor - combined - Nutanix Technologies India Pvt Ltd\nPune\nIN ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com URL:https://hasgeek.com/fifthelephant/2026-pune/schedule/agentic-triage-en gine-AQnqnC8W4ZgwfXywonRmHj BEGIN:VALARM ACTION:display DESCRIPTION:Sponsored talk: Agentic triage engine in Ursa Major and Ursa M inor - combined in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT BEGIN:VEVENT SUMMARY:AIGrants.in - mission and how you can participate DTSTART:20260228T071000Z DTEND:20260228T072000Z DTSTAMP:20260421T120126Z UID:session/KUsmN1gJQ9vDyUfErRvcJY@hasgeek.com SEQUENCE:1 CREATED:20260228T062506Z GEO:18.571009876468388;73.77414356162001 LAST-MODIFIED:20260228T062509Z LOCATION:Ursa Major and Ursa Minor - combined - Nutanix Technologies India Pvt Ltd\nPune\nIN ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com BEGIN:VALARM ACTION:display DESCRIPTION:AIGrants.in - mission and how you can participate in Ursa Majo r and Ursa Minor - combined in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT BEGIN:VEVENT SUMMARY:Birds of Feather (BOF) session: The ostrich in your war rooms: BI bottlenecks DTSTART:20260228T071500Z DTEND:20260228T081500Z DTSTAMP:20260421T120126Z UID:session/FQf4rZ8cAdTZoFXPQnUsUL@hasgeek.com SEQUENCE:29 CREATED:20260203T121136Z GEO:18.571009876468388;73.77414356162001 LAST-MODIFIED:20260228T062513Z LOCATION:Alpha Century room for BOFs - Nutanix Technologies India Pvt Ltd\ nPune\nIN ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com BEGIN:VALARM ACTION:display DESCRIPTION:Birds of Feather (BOF) session: The ostrich in your war rooms: BI bottlenecks in Alpha Century room for BOFs in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT BEGIN:VEVENT SUMMARY:Building the systems that build the software DTSTART:20260228T072000Z DTEND:20260228T075000Z DTSTAMP:20260421T120126Z UID:session/NkdvgggYBDTYch565JHQDi@hasgeek.com SEQUENCE:11 CATEGORIES:Track 1 AI in Software Development Life Cycle (SDLC),30 mins ta lk CREATED:20260203T121714Z DESCRIPTION:Other possible talk titles:\n\n- What Do Engineers Do When the Machine Writes the Code?\n- Learning to Work with Something That Almost W orks\n- Old disciplines for new machines\n- Mise en place pour machines ca pricieuses\n\nMechanical engineers aren’t assemblers—they set up facto ries. Civil engineers aren’t construction workers—construction compani es do the construction. Similarly\, creating software should go beyond pro ducing code—we should cultivate the sociotechnical system that produces the code. While building *StoryMachine*\, our experimental open source too l that helps PMs and Engineers cut out “common sense” units of work\, we have uncovered a lot more about what it means to build software with on -demand\, stochastic and jagged machine intelligence.\n\nIn this session I ’ll talk about how to refactor your engineering team to accommodate this new reality. I’ll make a case for:\n\n- Tightening the collaborative lo op between engineering and product—what this looks like in practice when you have AI involved (”Taste and Adjust”/”The value is in the conve rsation”)\n- How Sutton’s bitter lesson calls for radical simplicity a nd *mise en place* thinking (don’t fight the chef\, just set up their ki tchen)\n- Why scientific thinking and basic statistical literacy is table stakes in a world where we have to deal with stochastic outputs from AI (h ow to avoid data and benchmarks leading you astray)\n\nAll of this will be backed up by our experience building and evaluating *StoryMachine*—alon g with a decade’s worth of lessons from deploying complex software engin eering projects at nilenso.\n\n## Takeaways\n\n- The main takeaway will be our high-confidence recommendations for all the new skills that Software Engineers and Engineering Leaders need to learn in order to work in a futu re dominated by use of machine intelligence.\n- Specifically\, I will cove r this triad: the essence of iterative refinement (Agile\, OODA whatever y ou want to call it) and why it is more powerful today\, how to design an e nvironment that gets out of the way of AI agents and how to adapt your thi nking to account for risk and uncertainty—and why this triad is actually nothing new at all.\n\n## References\nLinks that contain about some of th e ideas I'll bring up in the talk. (actual slides to come soon)\n\n[Taste and Adjust](https://blog.nilenso.com/blog/2025/11/26/how-to-work-with-prod uct-taste-and-adjust/)\n\n[Minimum Viable Benchmark](https://blog.nilenso. com/blog/2025/11/28/minimum-viable-benchmark/)\n\n[Artisanal shims for the bitter lesson age](https://blog.nilenso.com/blog/2025/10/14/bitter-lesson -applied-ai/)\n\n[The quality of AI-assisted software depends on unit of w ork management](https://blog.nilenso.com/blog/2025/09/15/ai-unit-of-work/) \n\n## Bio\n\n[Atharva](https://atharvaraykar.com/about/)\, a member of [n ilenso](https://www.notion.so/FifthEl-Pune-CFP-2c30f0425dae80a08010c22f390 98a0a?pvs=21)\, has been tinkering with LLMs to figure out how to build pr oducts that deliver on the hype in production.\n\nHis first exposure to se rious software development was with the Git project\, where he rewrote par ts of the submodule functionality by emailing patches to the maintainer. S ince then\, he’s been fascinated with the sociotechnical dynamics of bui lding software.\n\nWhile at nilenso\, he helped a hyperlocal delivery star tup revamp their payout systems\, solved data integration challenges at a non-profit building population-scale software and helped build voice AI ag ents in production. GEO:18.571009876468388;73.77414356162001 LAST-MODIFIED:20260208T101517Z LOCATION:Ursa Major and Ursa Minor - combined - Nutanix Technologies India Pvt Ltd\nPune\nIN ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com URL:https://hasgeek.com/fifthelephant/2026-pune/schedule/building-the-syst ems-that-build-the-software-NkdvgggYBDTYch565JHQDi BEGIN:VALARM ACTION:display DESCRIPTION:Building the systems that build the software in Ursa Major and Ursa Minor - combined in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT BEGIN:VEVENT SUMMARY:Lunch break DTSTART:20260228T075000Z DTEND:20260228T084500Z DTSTAMP:20260421T120126Z UID:session/AJ8zzrcPeeLhxCkRWgZLWn@hasgeek.com SEQUENCE:12 CREATED:20260203T121207Z GEO:18.571009876468388;73.77414356162001 LAST-MODIFIED:20260208T064604Z LOCATION:Ursa Major and Ursa Minor - combined - Nutanix Technologies India Pvt Ltd\nPune\nIN ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com BEGIN:VALARM ACTION:display DESCRIPTION:Lunch break in Ursa Major and Ursa Minor - combined in 5 minut es TRIGGER:-PT5M END:VALARM END:VEVENT BEGIN:VEVENT SUMMARY:Transforming robot and vehicle autonomy with end-to-end AI stacks and World Foundation Models (WFMs) DTSTART:20260228T084500Z DTEND:20260228T092000Z DTSTAMP:20260421T120126Z UID:session/TwK3pLYcNKAB4HYeJuBRzW@hasgeek.com SEQUENCE:18 CATEGORIES:Track 2: AI in Manufacturing & Digital Transformation,30 mins t alk CREATED:20260203T121233Z DESCRIPTION:The landscape of autonomous systems is rapidly evolving\, shif ting from traditional\, modular approaches to integrated\, end-to-end AI s tacks. These stacks represent a significant leap forward\, capable of dire ctly translating raw perception inputs – such as camera images and lidar data – into precise action control outputs without relying on intermedi ate\, hand-engineered modules. This streamlined architecture promises incr eased efficiency and adaptability\, but hinges on the availability of mass ive\, high-quality datasets for effective training.\n\nOur work concentrat es on enabling these end-to-end AI stacks through a deep dive into trainin g and development methodologies utilizing synthetic data generated by Worl d Foundation Models (WFMs). We explore how to leverage the vast potential of WFMs to create realistic and diverse large-scale datasets necessary to train and validate these complex AI systems. This approach addresses sever al critical challenges in the autonomy space\, paving the way for more rob ust and capable self-driving technologies.\n\n### Key Takeaways:\n- **Shif t to End-to-End AI:** The autonomy field is moving towards end-to-end AI s tacks that directly link perception to action.\n- **WFMs Enable Synthetic Data:** World Foundation Models are a key enabler for generating realistic and diverse synthetic data for training and validation.\n\n### Target Aud ience:\n- **AI/ML Engineers (Autonomy Focus):** Professionals specializing in artificial intelligence and machine learning\, specifically those work ing on autonomous systems.\n- **Research Scientists (Robotics/AI):** Resea rchers exploring the latest advancements in robotics\, AI\, and autonomous systems.\n- **Technical Leaders (Automotive/Robotics):** Individuals in l eadership positions responsible for strategic technology decisions within the automotive or robotics industries.\n\n### About Speaker:\nAbhilash SK is a Technical Architect at the Autonomous Driving practice in KPIT. His r esponsibilities include leading techincal research for autonomous driving domain with a focus on E2E AI stacks and WFMs. He holds a Masters Degree f rom Bangalore University and has a wide range of publications and patents on AI and related topics. GEO:18.571009876468388;73.77414356162001 LAST-MODIFIED:20260222T131015Z LOCATION:Ursa Major and Ursa Minor - combined - Nutanix Technologies India Pvt Ltd\nPune\nIN ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com URL:https://hasgeek.com/fifthelephant/2026-pune/schedule/transforming-robo t-and-vehicle-autonomy-with-end-to-end-ai-stacks-and-world-foundation-mode ls-TwK3pLYcNKAB4HYeJuBRzW BEGIN:VALARM ACTION:display DESCRIPTION:Transforming robot and vehicle autonomy with end-to-end AI sta cks and World Foundation Models (WFMs) in Ursa Major and Ursa Minor - comb ined in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT BEGIN:VEVENT SUMMARY:Birds of Feather (BOF) session: Best Practices for successful AI a doption in SDLC DTSTART:20260228T085000Z DTEND:20260228T095000Z DTSTAMP:20260421T120126Z UID:session/MGhgzbo6DuZaT1UcP2URJ4@hasgeek.com SEQUENCE:5 CATEGORIES:Track 2: AI in Manufacturing & Digital Transformation,30 mins t alk CREATED:20260222T130715Z DESCRIPTION:\n# BoF Propsal: Best Practices for successful AI adoption in SDLC\nAI is changing software development in two ways at once: we’re shi pping AI-powered features that used to be research\, and we’re adopting coding agents that can generate code faster than teams can comfortably rev iew. The opportunity is huge—but without mature engineering systems\, th e result is predictable: impressive demos\, fragile production\, and a gro wing gap between how fast we can change code and how safely we should.\n\n This BoF is built around a simple idea: the practices that make AI agents reliable in production are the same practices that make coding agents work for you instead of against you. The “boring” parts of engineering—d ocumentation\, specs\, tests\, CI/CD\, code reviews\, observability—are not legacy rituals. They’re the control system that lets you harness AI acceleration without turning your SDLC into chaos.\n\nIn this discussion\, we’ll share best practices and patterns that have worked in real teams: \n • Mature engineering systems & gates that scale with AI (review rubri cs\, CI policies\, eval harnesses\, rollout guardrails)\n • Bridging dem os → production for AI features (telemetry-first design\, failure modes\ , fallbacks\, canaries/shadow mode)\n • Agentic coding loops that are sa fe and high-leverage (small diffs\, acceptance criteria\, automated verifi cation\, human-in-the-loop approvals)\n\n**This session is relevant if you ’re**:\n • Shipping AI features in an existing product and want them t o be reliable\, observable\, and maintainable\n • Introducing coding age nts into your team and want speed without regressions\n • Using AI-assis ted coding personally and want a workflow that consistently produces merge able\, production-grade changes\n\n**Expected takeaways**\n • A practica l “minimum viable maturity” checklist for AI adoption in the SDLC\n • A shared set of patterns for reliable AI features and productive codin g agents—because they’re more similar than they look\n\n----\n----\n-- --\n\n*Previous talk submission*\n\nThis talk starts with a familiar story : the 8:30 PM scramble before a demo\, endlessly tweaking prompts until th e bot behaves just enough for tomorrow morning. The demo goes well\, every one is happy\, the feature is greenlit—and then it quietly falls apart i n production. Users repeat themselves. Interruptions break the flow. Tool calls misfire. You have recordings but no traces\, complaints but no repro steps\, and you’re stuck in the same “tweak and pray” loop—just w ith more traffic and higher stakes.\n\nIn this session\, I’ll argue that the difference between “cool demo” and “reliable product” is not model choice or prompt cleverness\, but engineering maturity: documentatio n\, observability\, evals\, datasets\, CI/CD\, and feedback loops. We’ll reframe AI product development as discovery\, not invention and walk thro ugh concrete practices for building that discovery engine: how to log and trace every LLM and tool call\, design evals that actually catch regressio ns\, turn production traffic into datasets\, and build a flywheel where ev ery failure makes the system stronger. You’ll leave with a pragmatic che cklist you can apply to your current AI project without a full platform re write.\n\n⸻\n\nMention 1–2 takeaways from your session\n • You’ll learn a practical definition of engineering maturity for AI applications a nd a minimal set of non-negotiables (docs\, observability\, evals\, datase ts\, CI/CD) that turn fragile demos into reliable systems.\n • You’ll leave with a concrete “flywheel” pattern for AI products—how to capt ure data\, tag outcomes\, run evals\, and iterate—so you can answer “C an we ship this to 100\,000 users?” with data instead of hope.\n\n⸻\n\ nWhich audiences is your session going to be beneficial for?\n\nThis sessi on will be most useful for:\n • Engineering managers and tech leads resp onsible for shipping AI features to production\n • Senior/principal engi neers and ML/AI engineers working with LLMs\, tools\, and agents\n • Pro duct managers and founders trying to turn promising AI prototypes into rel iable products\n • Platform / infra / DevOps engineers designing interna l AI platforms or evaluation/observability stacks\n\n⸻\n\nAdd your bio – who you are\; where you work\n\nI’m Govind Joshi\, an independent so ftware engineer based in India who spends an unreasonable amount of time b uilding AI-powered systems that actually have to work in the real world. I focus on applied AI: LLM-driven agents that can call tools\, handle real users over phone and chat\, and operate reliably under production traffic. \n\nOver the last few years\, I’ve worked with teams to design and ship AI assistants\, voice bots\, and evaluation/observability pipelines for LL M applications. I care a lot about the “boring” parts—architecture\, evals\, monitoring\, and engineering maturity—and how they turn AI demo s into products you can trust. GEO:18.571009876468388;73.77414356162001 LAST-MODIFIED:20260222T131318Z LOCATION:Alpha Century room for BOFs - Nutanix Technologies India Pvt Ltd\ nPune\nIN ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com URL:https://hasgeek.com/fifthelephant/2026-pune/schedule/engineering-matur ity-is-all-you-need-MGhgzbo6DuZaT1UcP2URJ4 BEGIN:VALARM ACTION:display DESCRIPTION:Birds of Feather (BOF) session: Best Practices for successful AI adoption in SDLC in Alpha Century room for BOFs in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT BEGIN:VEVENT SUMMARY:Ship faster: architecting AI-driven test automation in production DTSTART:20260228T092500Z DTEND:20260228T100000Z DTSTAMP:20260421T120126Z UID:session/XcTPgEWNxJ189KgDJzgsHj@hasgeek.com SEQUENCE:16 CATEGORIES:Track 1 AI in Software Development Life Cycle (SDLC),30 mins ta lk CREATED:20260203T121729Z DESCRIPTION:## Description\nTraditional test automation breaks down at sca le: test suites become brittle\, coverage lags behind feature velocity\, a nd QA teams spend more time maintaining tests than finding real issues. At IDfy\, we ran into similar problems while shipping multiple times a week across APIs\, UIs\, and ML-based solutions. This session walks through how we designed and built an AI-powered QA automation platform that fundament ally rethinks test creation\, execution\, and analysis. By using LLMs to u nderstand PRDs\, visual designs (screenshots\, PDFs\, videos)\, and API sp ecifications\, we automated the generation of comprehensive test cases cov ering positive\, negative\, and edge scenarios with minimal QA overhead.\n \nThe talk goes deep into the architecture and practical implementation of a unified platform that supports API\, UI\, and ML test automation at dif ferent stages of product maturity and scale (0-1\, 1-10\, and 10-100). You ’ll see how we approached context design and prompt strategies\, optimiz ed test execution with intelligent test selection\, and how we integrated the platform into our development lifecycle. The session will include a li ve demos of our internal tool\, some real production metrics (time and cos t savings)\, and concrete patterns you can adapt immediately in your own t eams.\n\n## Takeaways\n- How to design and implement an AI-driven test aut omation platform that dramatically improves coverage\, speed\, and QA prod uctivity across APIs and UIs.\n- Practical techniques for using LLMs to ge nerate\, execute\, and analyze tests from PRDs\, visuals\, and API specs: without rewriting your entire QA stack.\n\n## Audience\n- Audience Level: all levels\n- This session is especially valuable for QA engineers\, SDETs \, test automation architects\, engineering managers\, and product teams r esponsible for quality at scale. It’s also relevant for senior developer s and tech leads exploring how to apply LLMs pragmatically within developm ent lifecycles and developer workflows.\n\n## About the speaker\nManas Cha turvedi is a Tech Lead at IDfy. He has worked on building and maintaining multiple platforms at IDfy and worked on integrating AI into various syste ms and workflow GEO:18.571009876468388;73.77414356162001 LAST-MODIFIED:20260222T131035Z LOCATION:Ursa Major and Ursa Minor - combined - Nutanix Technologies India Pvt Ltd\nPune\nIN ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com URL:https://hasgeek.com/fifthelephant/2026-pune/schedule/accelerating-deve lopment-through-ai-powered-test-automation-a-real-world-journey-XcTPgEWNxJ 189KgDJzgsHj BEGIN:VALARM ACTION:display DESCRIPTION:Ship faster: architecting AI-driven test automation in product ion in Ursa Major and Ursa Minor - combined in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT BEGIN:VEVENT SUMMARY:Flash talks DTSTART:20260228T095500Z DTEND:20260228T103500Z DTSTAMP:20260421T120126Z UID:session/LRTPeFdmTnbSsWeNqMQBh8@hasgeek.com SEQUENCE:23 CREATED:20260203T124123Z GEO:18.571009876468388;73.77414356162001 LAST-MODIFIED:20260208T065707Z LOCATION:Alpha Century room for BOFs - Nutanix Technologies India Pvt Ltd\ nPune\nIN ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com BEGIN:VALARM ACTION:display DESCRIPTION:Flash talks in Alpha Century room for BOFs in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT BEGIN:VEVENT SUMMARY:Transforming ecommerce item catalogue stewardship using GenAI and vision computing DTSTART:20260228T100500Z DTEND:20260228T103500Z DTSTAMP:20260421T120126Z UID:session/NBe8roLWPDaHiPKLgH4cxd@hasgeek.com SEQUENCE:12 CATEGORIES:Track 2: AI in Manufacturing & Digital Transformation,30 mins t alk CREATED:20260203T121258Z DESCRIPTION:GenAI and Vision Computing models offer a way to simplify and automate ecommerce item catalogue stewardship tasks. They are capable of e verything between from predicting and enriching attributes for better gues t experience and findability to creating automated infographics that tell the story of the item in one picture. This powers the e-commerce platform by increasing data fidelity and providing richer data for customers across millions of SKUs and hundreds of item types. Gen AI and vision computing are transforming item onboarding and management processes and is having a deep impact on the future of catalogue management.\n\nIn this talk\, we wi ll see share how Target (a $100B+ retailer based out of US) is architectin g the future of catalogue management and rewriting/simplifying the process using GenAI and Vision Models.\n\nListeners will be able to takeaway - \n - How GenAI and Vision computing is disrupting the traditional way of item catalogue managment\n- How it is enabling creation of richer content type for items\n- Pre-requisites and challenges in architecting such a system\ n\nThis is going to be beneficial for:\n- Technologists\n- Ecommerce busin ess leaders and participants\n- AI enthusiasts who want to see cutting edg e application of AI\n\nBio:\nMudit is a seasoned scientist who works as a Senior Director\, Data Sciences at Target. Target is one of the largest ec ommerce retailer and retail media network in the United States. Mudit lead s a team of data scientists and engineers to embed intelligent decisioning at scale to optimize and improve retail media performance and automate an d enrich e-commerce catalogue over millions of SKUs. GEO:18.571009876468388;73.77414356162001 LAST-MODIFIED:20260222T131048Z LOCATION:Ursa Major and Ursa Minor - combined - Nutanix Technologies India Pvt Ltd\nPune\nIN ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com URL:https://hasgeek.com/fifthelephant/2026-pune/schedule/transforming-ecom merce-item-catalogue-stewardship-using-genai-and-vision-computing-NBe8roLW PDaHiPKLgH4cxd BEGIN:VALARM ACTION:display DESCRIPTION:Transforming ecommerce item catalogue stewardship using GenAI and vision computing in Ursa Major and Ursa Minor - combined in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT BEGIN:VEVENT SUMMARY:Evening break DTSTART:20260228T103500Z DTEND:20260228T110000Z DTSTAMP:20260421T120126Z UID:session/JrtQXS4ggTHXrce5W8sFn2@hasgeek.com SEQUENCE:2 CREATED:20260208T065039Z GEO:18.571009876468388;73.77414356162001 LAST-MODIFIED:20260208T065047Z LOCATION:Ursa Major and Ursa Minor - combined - Nutanix Technologies India Pvt Ltd\nPune\nIN ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com BEGIN:VALARM ACTION:display DESCRIPTION:Evening break in Ursa Major and Ursa Minor - combined in 5 min utes TRIGGER:-PT5M END:VALARM END:VEVENT BEGIN:VEVENT SUMMARY:Programmable knowledge for AI native SDLC DTSTART:20260228T110000Z DTEND:20260228T113500Z DTSTAMP:20260421T120126Z UID:session/EgMDcRARjDf59hBeBUzUZN@hasgeek.com SEQUENCE:13 CATEGORIES:Track 1 AI in Software Development Life Cycle (SDLC),30 mins ta lk CREATED:20260203T121827Z DESCRIPTION:Databases gave us programmable data. REST APIs gave us program mable information. With LLMs we are poised for programmable knowledge. But when it comes to complex domains\, are we ready?\n\nWhen it comes to syst ems like a production software stack\, knowledge is held not just in the r aw data\, but in the entities of the system and their relationships. While information can be represented as a vector\, knowledge is essentially a g raph. In this talk I will present techniques that have proven in the real world to boost quality and accuracy of agents\, by augmenting them with pr ogrammable knowledge graphs.\n\nWe will look at the parts of a programmabl e knowledge system\, with a real-world case study.\n\nThe audience can exp ect to learn about practical challenges in using LLMs for complex domains\ , and gain at least a conceptual understanding of the components of a prog rammable knowledge agentic system.\n\nAditya Godbole is the CTO of last9.i o. He has seen the twists and turns of the software industry for 24 years and has worked across technologies and domains from embedded systems using assembly\, to ML models on the cloud.\nAlong with building software produ cts and companies\, he is also a student of physics\, mathematics and tenn is. GEO:18.571009876468388;73.77414356162001 LAST-MODIFIED:20260222T131055Z LOCATION:Ursa Major and Ursa Minor - combined - Nutanix Technologies India Pvt Ltd\nPune\nIN ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com URL:https://hasgeek.com/fifthelephant/2026-pune/schedule/programmable-know ledge-for-ai-native-sdlc-EgMDcRARjDf59hBeBUzUZN BEGIN:VALARM ACTION:display DESCRIPTION:Programmable knowledge for AI native SDLC in Ursa Major and Ur sa Minor - combined in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT BEGIN:VEVENT SUMMARY:Birds of Feather (BOF) session: Quality Assurance (QA): AI strateg ies for faster\, reliable testing DTSTART:20260228T111500Z DTEND:20260228T121500Z DTSTAMP:20260421T120126Z UID:session/KCYJfnhkTjXox51vFnQF8f@hasgeek.com SEQUENCE:20 CREATED:20260203T124206Z GEO:18.571009876468388;73.77414356162001 LAST-MODIFIED:20260223T054205Z LOCATION:Alpha Century room for BOFs - Nutanix Technologies India Pvt Ltd\ nPune\nIN ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com BEGIN:VALARM ACTION:display DESCRIPTION:Birds of Feather (BOF) session: Quality Assurance (QA): AI str ategies for faster\, reliable testing in Alpha Century room for BOFs in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT BEGIN:VEVENT SUMMARY:Smart agents\, smarter schedules: Agentic AI for complex optimizat ion in airlines DTSTART:20260228T114000Z DTEND:20260228T121500Z DTSTAMP:20260421T120126Z UID:session/DeLoNWDPDFEYfrdYi8CJSQ@hasgeek.com SEQUENCE:11 CATEGORIES:Track 2: AI in Manufacturing & Digital Transformation,30 mins t alk CREATED:20260203T121326Z DESCRIPTION:Extended Abstract:-\n\n**Constrained optimization** lies at th e heart of many complex systems\, where multiple limited resources must be allocated efficiently to achieve specific goals\, while satisfying strict operational constraints. These problems generally sit at the core of **lo gistics optimisation**\, **smart scheduling**\, etc. They are often high-d imensional\, dynamic\, and involve tightly coupled dependencies between va riables. Traditional methods like Mixed-Integer Linear Programming or Heur istics often struggle in dynamic\, distributed settings due to their centr alized and rigid nature. Agentic AI offers a compelling alternative by mod eling system components as autonomous agents capable of local reasoning\, coordination\, and negotiation.\n\nThis talk illustrates the power of **Ag entic AI for Airline scheduling**\, which is an exceptionally complex\, re al-world example of constrained optimization. Here\, aircraft\, crew\, fue l resources\, etc. must be jointly scheduled under tight timing constraint s\, regulatory rules\, and disruption risks. By representing multiple elem ents as intelligent agents\, agentic AI enables decentralized decision-mak ing and dynamic re-optimization. The talk will demonstrate how integrating agentic models with classical optimization solvers creates hybrid framewo rks that are not only scalable but also resilient to real-time changes.\n\ nKey takeaways:-\n- Learn how agentic AI enables decentralized\, adaptive\ , optimal solutions\n- See the complexity involved in optimisations like a irline scheduling \n- See how agentic AI solves such complex problems\n\nT arget audience:-\n- Practitioners keen to learn amazing potential of agent ic AI\n- Data science and AI enthusiasts\n- Folks looking to solve complex optimization problems\n\nSpeaker bio:- \nDr. Karthika Vijayan is a Soluti on Consultant at Sahaj Software. She has been conducting research in the f ield of conversational AI with voice and text data for almost a decade. He r research has been published in several journals and presented at various international conferences. Prior to joining Sahaj Software\, she worked a s a research fellow at the National University of Singapore and at IISc Ba ngalore. She has done her PhD from IIT Hyderabad.\n\nPrevious talk links:- \n- https://www.youtube.com/watch?v=o6YHcDLod8A\n- https://www.youtube.co m/watch?v=-uoUwGpzIL0\n- https://www.youtube.com/watch?v=kphYc_lvKIk&list= PLkPaq00oPRfzz9O4q06rOL2dHCEX7PQwU&index=18\n- https://www.youtube.com/wat ch?v=gvJhtBdmUi8&t=897s\n\nProfile links:-\n- https://scholar.google.com/c itations?user=fJp6O0UAAAAJ&hl=en\n- https://www.linkedin.com/in/karthika-v ijayan/\n- https://www.researchgate.net/profile/Karthika-Vijayan\n\n\n\n\n GEO:18.571009876468388;73.77414356162001 LAST-MODIFIED:20260222T131214Z LOCATION:Ursa Major and Ursa Minor - combined - Nutanix Technologies India Pvt Ltd\nPune\nIN ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com URL:https://hasgeek.com/fifthelephant/2026-pune/schedule/smart-agents-smar ter-schedules-agentic-ai-for-complex-optimization-DeLoNWDPDFEYfrdYi8CJSQ BEGIN:VALARM ACTION:display DESCRIPTION:Smart agents\, smarter schedules: Agentic AI for complex optim ization in airlines in Ursa Major and Ursa Minor - combined in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT BEGIN:VEVENT SUMMARY:Wrap-up & feedback DTSTART:20260228T121500Z DTEND:20260228T123000Z DTSTAMP:20260421T120126Z UID:session/JmBULaG4RTmA3Hy4aK1q33@hasgeek.com SEQUENCE:8 CREATED:20260203T121427Z GEO:18.571009876468388;73.77414356162001 LAST-MODIFIED:20260208T065300Z LOCATION:Ursa Major and Ursa Minor - combined - Nutanix Technologies India Pvt Ltd\nPune\nIN ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com BEGIN:VALARM ACTION:display DESCRIPTION:Wrap-up & feedback in Ursa Major and Ursa Minor - combined in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT END:VCALENDAR