This page is only for submissions for The Fifth Elephant 2026 annual conference

Dates

• Conference on Friday, 31 July, at the NIMHANS Convention Centre
• Workshops in July (7, 17, 25) and Saturday, 1 August in Marathalli, Koramangala & Whitefield

To attend the conference, get an annual membership - https://hasgeek.com/fifthelephant#memberships

How to submit

1. Submit your abstract. Your abstract should clearly describe:

   • the problem or topic being addressed,
   • why it is relevant,
   • who the session is intended for, and
   • the key takeaways attendees can expect.

2. Add a link to draft slides within 3–5 days - after submitting your abstract, add a link to the draft version of your slides within 3–5 days. Draft slides help editors better evaluate the structure, depth, and delivery of the session.
   Submissions without draft slides may not be reviewed or may experience delays in receiving feedback.

3. Add a 2-minute elevator pitch video - introducing your session and demonstrating your speaking style. This helps the editorial team assess presentation clarity and audience engagement.

4. Make your submission here - https://hasgeek.com/fifthelephant/fifthelephant-2026-call-for-submissions/sub

Review timelines and submission deadlines

• Feedback via comments may be shared early for submissions that include slide links.
• Selected speakers will be contacted starting 17 June onwards.
• The final submission deadline is 25 June 2026.
• Not every submission may fit the annual conference schedule in July. Some proposals may be considered for future community sessions, including weekly reviews and monthly meet-ups.

Conference tracks

Track 1: Data Engineering & Infrastructure

The hard problems are just beginning.

The orchestration tools have matured. The primitives are no longer the argument.

When agentic systems generate ad hoc queries that shatter your partition assumptions, when a pipeline written by an AI model goes to production with your name on the approval, when an autonomous quality monitor acts on a data quality miss before a human sees it, and when your cloud bill arrives with a line item no one planned for.

This track is built around a single mandate: real systems under real production constraints : latency, cost, and scale. Track editors are looking for submissions that reflect what’s actually working (and what isn’t) in production data systems, especially as AI workloads reshape infrastructure assumptions.

We want the story of what broke, what you threw away, and what you rebuilt to survive.

AI-native data systems

Who is consuming your data; humans or agents? When it’s agents, the engineering decisions change fundamentally. Token budget replaces query cost. Descriptions of relationships matter more than the relationships themselves. Functions beat tables. Logs beat docs.

We want talks on:

• Data engineering for foundational model training and fine-tuning
• Data pipelines for inference and post-training steps
• Data stores and retrieval patterns for GenAI and LLM-based applications
• Memory management and agent-to-agent communication
• Evals, guardrails, and observability implemented in production systems

Foundations & storage

The format wars produced two survivors worth betting on. The catalog and query engine landscape did not stand still. And the assumption that you need a distributed cluster for serious workloads is being quietly dismantled.

We want talks on:

• Lakehouse and lakebase architectures: what’s working in production
• Table format evolution: Iceberg, DuckLake, and how metadata management is changing
• Query engines in the wild: including newer entrants like Apache DataFusion
• In-memory and local databases: when and why they make sense
• Realtime CDC and streaming data patterns

Governance, compliance & data quality

Governance frameworks that exist only in documentation are not what we are looking for. The metadata layer has moved from nice-to-have to load-bearing infrastructure; and agents hallucinate metrics without a governed definition layer. Meanwhile, DPDP is no longer a future problem.

We want talks on:

• Practical approaches to PII detection, masking, and management at scale
• Metadata management strategies that actually hold up in large organisations
• Governance frameworks adapted for AI and LLM pipelines
• Data quality practices in the context of model inputs and outputs

Ops, reliability & costs

Cloud bills are bleeding, and the GPU line item has arrived. Embedding and vector compute are now cost centres that data engineers own, not ML platform teams. And in closed-loop systems where agents act on data before a human sees it, standard observability pillars are no longer enough.

We want talks on:

• Observability practices for data and AI infrastructure
• Agent incident response and SRE patterns for agentic systems
• Cost optimisation strategies for AI infrastructure: compute, storage, egress
• Hardware and compute considerations; sub-architecture patterns for inference and training workloads

Orchestration & pipelines

The engineer who gave an agent write access to production has a different relationship with orchestration than the one who runs nightly batch jobs. Migration stories, operational failure modes, and the integration of agent workflows into pipeline tooling.

We want talks on:

• Production use cases with tools like Airbyte, Fivetran, Dagster, Prefect, or Temporal
• Patterns for integrating orchestration with AI and agent workflows
• Migration stories from legacy pipelines to modern orchestration
• Operational lessons from running pipelines at scale

A standing invitation to disagree

We are holding a slot for the talk that pushes back on everything above.

Most companies are not running agentic data systems. Most data engineering is still ETL, warehousing, and batch processing; unglamorous, unfinished, and real. If your strongest opinion is “your team doesn’t need agents, it needs a working warehouse and tested pipelines”. You can back it with production experience. It keeps the rest of the programme honest.

Track 2: Building and Implementing AI Tools & Agents in Production

The demo always works. Production is where every assumption breaks.

Every team has now built an agent. Most of them work brilliantly in the demo. A smaller number are actually in production. A smaller number still are in production and behaving as designed six months later.

The gap between those groups is not a model problem or a prompt problem. It is testing reliability under real load, observability when the system makes a decision you didn’t anticipate, cost when the token bill arrives, and accountability when an agent acts on bad input before a human notices.

This track is for the engineers and teams who have crossed into production, or are close enough to see what’s waiting for them there. We want the case study where the architecture changed after the first real incident. The integration that took three attempts to get right. The evaluation strategy that replaced gut feel. The business case that survived contact with a CFO

Building production-ready agents and tools

Building one that holds up under real usage, degrades gracefully, and doesn’t surprise you at 2am is a different discipline entirely. The MCP ecosystem has matured enough to have opinions about; and enough production scar tissue to share.

We want talks on:

• Building production-ready agent frameworks and tools
• Implementing MCP (Model Context Protocol) servers and clients
• Tool development for specific domains: customer support, operations, data analysis
• Designing tool interfaces for LLM consumption
• Creating reusable agent components and libraries

Multi-agent orchestration

When one agent calls another, you have distributed systems problems again; latency, partial failure, state management, and the unique joy of debugging a chain where every step was technically successful and the output is still wrong. Orchestration at this layer is genuinely unsolved, and practitioners are working it out in production.

We want talks on:

• Multi-agent orchestration patterns that held up under real conditions
• Scaling agent-based systems: where the architecture broke and replacements have to be done
• Reliability and monitoring for agentic systems
• Operational challenges and solutions from running agents at scale

Agents in the enterprise

Enterprise environments bring SSO, procurement, compliance reviews, change management, and colleagues who did not ask for an AI agent in their workflow. The integration stories from teams who navigated all of that are the ones this audience needs.

We want talks on:

• Real-world case studies of agents in production: what changed between v1 and what runs today
• Integration patterns for enterprise environments
• Migrating from traditional automation to AI agents; the technical and organisational friction
• ROI and business impact of agentic systems subject to what can be shared.
• Team workflows with AI agents; what changed for the people

Evaluation, testing & quality assurance

How do you write a test for a system that is non-deterministic by design? How do you know your agent is getting better when “better” is partly subjective? These are genuinely hard problems, and the field is developing real answers.

We want talks on:

• Agent evaluation strategies and quality assurance frameworks
• Testing strategies for non-deterministic systems
• Prompt engineering and agent optimisation; what actually moved the metric
• Debugging approaches when the failure mode is “the reasoning was plausible but wrong”

Observability, security & human-in-the-loop

When an agent takes an action in a closed loop, the standard observability stack tells you what happened. Security and human-oversight patterns are not add-ons to agent architecture; they determine whether the system is trustworthy enough to run unsupervised.

We want talks on:

• Observability and debugging for agentic systems in production
• Security and safety in agentic systems: enforcing limits that hold under adversarial conditions
• Human-in-the-loop patterns: where they are genuinely necessary and where they become a bottleneck
• Cost optimisation strategies: token budgets, caching, batching, and the real bill

Tool and framework selection

The framework landscape changes faster than most teams can evaluate it. The engineers who have actually run LangChain, LlamaIndex, CrewAI, Temporal, or a bespoke stack through a production incident have something to say that a benchmark table cannot. Honest trade-off analysis from someone who made the call and lived with it is the most useful thing in this space right now.

We want talks on:

• Tool choice and framework selection: how you decided, what you learned, what you’d change
• Migration stories: between frameworks, between approaches, or off a framework entirely
• Build vs. buy decisions with real context about what tipped the balance

A standing invitation to disagree

We are holding a slot for the talk that pushes back on everything above.

Most agent projects are still in pilot. Many will not make it to production. The honest talk about why poorly defined scope, unmeasurable success criteria, organisational readiness that was never there; is as valuable as any success story. If you shut down an agent project and learned something the field needs to hear, this stage is yours.

Workshop topics

Topics include, but are not limited to:

• MCP server creation with different protocol variations
• Agent frameworks like LangChain and LlamaIndex - which are widely used
• Implementing specific agent patterns (e.g., ReAct, chain-of-thought)
• Tool development tutorials for agentic systems
• Integration exercises with real APIs and services
• Hands-on prompt engineering and optimization techniques

Conference editors

• Jagadish K. (Tryft)
• Ramkrishna Reddy Yekulla (Red Hat)
• Ranganadh Thata (Mico)
• Sitaram Shelke (NVIDIA)
• Yash Gandhi (OrcaSheets)

Consulting editor

• Ravi Balgi (Datanimbus)

Got a question? Need help?

📞 Call or text The Fifth Elephant at (91) 7676332020
📧 Email info@hasgeek.com
💰 For sponsorship inquiries, email sales@hasgeek.com