BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//HasGeek//NONSGML Funnel//EN
DESCRIPTION:Open Source AI Meet-up - Bangalore edition
X-WR-CALDESC:Open Source AI Meet-up - Bangalore edition
NAME:Advancing multimodal and agentic AI: systems\, storage & scalability
X-WR-CALNAME:Advancing multimodal and agentic AI: systems\, storage & scal
 ability
REFRESH-INTERVAL;VALUE=DURATION:PT12H
SUMMARY:Advancing multimodal and agentic AI: systems\, storage & scalabili
 ty
TIMEZONE-ID:Asia/Kolkata
X-PUBLISHED-TTL:PT12H
X-WR-TIMEZONE:Asia/Kolkata
BEGIN:VEVENT
SUMMARY:Check-in - for participants (9th floor)
DTSTART:20250404T081500Z
DTEND:20250404T083000Z
DTSTAMP:20260421T120133Z
UID:session/Xkb2kEP5fj5bB4u9t5EB8t@hasgeek.com
SEQUENCE:3
CREATED:20250401T084238Z
GEO:12.9418972;77.6941002
LAST-MODIFIED:20250402T061445Z
LOCATION:9th Floor\, MERCURY BLOCK - Nutanix Technologies India Pvt Ltd\nB
 engaluru\nIN
ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com
BEGIN:VALARM
ACTION:display
DESCRIPTION:Check-in - for participants (9th floor) in 9th Floor\, MERCURY
  BLOCK in 5 minutes
TRIGGER:-PT5M
END:VALARM
END:VEVENT
BEGIN:VEVENT
SUMMARY:Introduction to the Fifthel-OSAI community and the meet-up
DTSTART:20250404T083000Z
DTEND:20250404T084000Z
DTSTAMP:20260421T120133Z
UID:session/NDnASF5uQmUuDw2QEFfghm@hasgeek.com
SEQUENCE:1
CREATED:20250401T084308Z
GEO:12.9418972;77.6941002
LAST-MODIFIED:20250401T084311Z
LOCATION:9th Floor\, MERCURY BLOCK - Nutanix Technologies India Pvt Ltd\nB
 engaluru\nIN
ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com
BEGIN:VALARM
ACTION:display
DESCRIPTION:Introduction to the Fifthel-OSAI community and the meet-up in 
 9th Floor\, MERCURY BLOCK in 5 minutes
TRIGGER:-PT5M
END:VALARM
END:VEVENT
BEGIN:VEVENT
SUMMARY:Supercharging LLM tuning: optimizing performance for faster traini
 ng cycles
DTSTART:20250404T084000Z
DTEND:20250404T090500Z
DTSTAMP:20260421T120133Z
UID:session/GHDemLvUpTDnoi7yHYpiUe@hasgeek.com
SEQUENCE:4
CATEGORIES:AI and costs - optimization\, efficiency\, learnings,Sponsored 
 talk (25-35 mins),Blr OSAI meetup in April 2025
CREATED:20250401T084946Z
DESCRIPTION:## Abstract\nGetting your first language model tuning up and r
 unning is great\, however\, in an enterprise\, you would always want to ex
 tract more value out of your expensive infrastructure by having more train
 ing cycles completed. This session presents various concepts to increase l
 anguage model tuning throughput. Covering topics (available in open source
 ) from usage of simple to complex knobs down the stack.\n\nOverview of con
 cepts for throughput enhancement\n\n* Training knobs. e.g. gradient checkp
 ointing + increased batch size etc\n\n* Faster implementations. e.g. repla
 ce attention module with flash attention\, padding free\n\n* Data sampling
  / collation for efficient load distribution. e.g. multi pack data sampler
 \n\n* Sparse techniques + increased batch size. e.g. LoRA\, Prompt Tuning\
 n\n* Parallelisms. e.g. data parallel techniques like distributed data par
 allel and fully sharded data parallel (FSDP)\n\n* Fast kernels. e.g. repla
 ce common operations such as cross entropy loss with triton kernel impleme
 ntations.\n\n* Torch compile. e.g. compile torch code to kernels\n\n## Tak
 eaways\n\nIn-depth conceptual awareness and understanding of various knobs
  pin pointing to specific stack/methods to maximize throughput. Audiences 
 should also be able to get a hang of application of these concepts. \n\n##
  Which audiences is your session going to beneficial for?\n\nTuning users 
 or AI professionals.\n\n## Bio\n\nI am Mehant Kammakomati. I work as a res
 earch software engineer at IBM Research - India. I work on language model 
 tuning capabilities aiming to give best tuning experience to our users.
GEO:12.9418972;77.6941002
LAST-MODIFIED:20250401T085459Z
LOCATION:9th Floor\, MERCURY BLOCK - Nutanix Technologies India Pvt Ltd\nB
 engaluru\nIN
ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com
URL:https://hasgeek.com/fifthelephant/osai-blr-meetup/schedule/enhancing-l
 anguage-model-tuning-throughput-GHDemLvUpTDnoi7yHYpiUe
BEGIN:VALARM
ACTION:display
DESCRIPTION:Supercharging LLM tuning: optimizing performance for faster tr
 aining cycles in 9th Floor\, MERCURY BLOCK in 5 minutes
TRIGGER:-PT5M
END:VALARM
END:VEVENT
BEGIN:VEVENT
SUMMARY:Transition
DTSTART:20250404T090500Z
DTEND:20250404T091000Z
DTSTAMP:20260421T120133Z
UID:session/4VDVktpeahHH8YeH6Ft73a@hasgeek.com
SEQUENCE:2
CREATED:20250401T084533Z
GEO:12.9418972;77.6941002
LAST-MODIFIED:20250401T085431Z
LOCATION:9th Floor\, MERCURY BLOCK - Nutanix Technologies India Pvt Ltd\nB
 engaluru\nIN
ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com
BEGIN:VALARM
ACTION:display
DESCRIPTION:Transition in 9th Floor\, MERCURY BLOCK in 5 minutes
TRIGGER:-PT5M
END:VALARM
END:VEVENT
BEGIN:VEVENT
SUMMARY:Efficient fine-tuning: interpreting LoRA\, intrinsic dimension\, a
 nd their connection
DTSTART:20250404T091000Z
DTEND:20250404T094000Z
DTSTAMP:20260421T120133Z
UID:session/Gw5oDygWX3HCWEHVAZyALH@hasgeek.com
SEQUENCE:5
CATEGORIES:Next generation architectures,Speaking at the Fifth Elephant 20
 25 Annual Conference,30 mins talk
CREATED:20250401T084519Z
DESCRIPTION:In this talk\, we will explore the growing need to fine-tune l
 arge pre-trained models for specialized tasks\, and the limitations of con
 ventional fine-tuning methods—especially their high computational and st
 orage costs. We begin with Parameter-Efficient Fine-Tuning (PEFT) techniqu
 es\, focusing on LoRA (Low-Rank Adaptation)\, an adapter-based approach th
 at enables efficient model adaptation by introducing a small number of tra
 inable parameters.\n\nThrough a hands-on implementation of LoRA in a multi
 layer perceptron (MLP) for a binary classification task\, we’ll cover ad
 apter insertion\, parameter configuration\, and the evaluation of paramete
 r efficiency. We’ll also discuss real-world workflows—like sharing mod
 els via Hugging Face Hub—and explore practical extensions such as Quanti
 zed-LoRA for reducing inference-time memory usage.\n\nBuilding on this fou
 ndation\, the talk transitions into the theory of Intrinsic Dimension (ID)
 —the hypothesis that neural networks\, despite their large size\, may re
 quire only a few effective directions to learn. Using random subspace trai
 ning\, we measure ID and analyze how models behave when learning is constr
 ained to low-dimensional subspaces. This leads to a key insight: LoRA’s 
 efficiency aligns closely with the principles of intrinsic dimension\, off
 ering a deeper theoretical understanding of why PEFT methods work.\n\nThis
  talk bridges the ideas explored in my four-part blog series\, which aims 
 to demystify PEFT\, LoRA\, and intrinsic dimension for a broader audience.
  These blogs gained good visibility — receiving positive traction on [/r
 / Machine learning](https://www.reddit.com/r/MachineLearning/comments/1hkn
 z26/d_fine_tuning_large_language_models/) and ranking 7th on [Hacker News]
 (https://news.ycombinator.com/item?id=42493871)\, where they stayed on the
  front page for a day. First two blogs on LoRA became the basis of a talk 
 I gave at [PyCon India 2024](https://www.linkedin.com/posts/preethi-sriniv
 asan-3a221915_thank-you-pycon-india-2024-for-the-opportunity-activity-7246
 124662907158529-M_qN/?utm_source=share&utm_medium=member_desktop).\n\n
GEO:12.9418972;77.6941002
LAST-MODIFIED:20250401T085529Z
LOCATION:9th Floor\, MERCURY BLOCK - Nutanix Technologies India Pvt Ltd\nB
 engaluru\nIN
ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com
URL:https://hasgeek.com/fifthelephant/osai-blr-meetup/schedule/fine-tuning
 -large-models-with-fewer-parameters-lora-and-intrinsic-dimension-explained
 -Gw5oDygWX3HCWEHVAZyALH
BEGIN:VALARM
ACTION:display
DESCRIPTION:Efficient fine-tuning: interpreting LoRA\, intrinsic dimension
 \, and their connection in 9th Floor\, MERCURY BLOCK in 5 minutes
TRIGGER:-PT5M
END:VALARM
END:VEVENT
BEGIN:VEVENT
SUMMARY:Transition
DTSTART:20250404T094000Z
DTEND:20250404T094500Z
DTSTAMP:20260421T120133Z
UID:session/EjmdHW3nhkHDVEB36aeuXX@hasgeek.com
SEQUENCE:2
CREATED:20250401T085028Z
GEO:12.9418972;77.6941002
LAST-MODIFIED:20250401T085535Z
LOCATION:9th Floor\, MERCURY BLOCK - Nutanix Technologies India Pvt Ltd\nB
 engaluru\nIN
ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com
BEGIN:VALARM
ACTION:display
DESCRIPTION:Transition in 9th Floor\, MERCURY BLOCK in 5 minutes
TRIGGER:-PT5M
END:VALARM
END:VEVENT
BEGIN:VEVENT
SUMMARY:Building Bharat’s LLMs: speech & text for a billion
DTSTART:20250404T094500Z
DTEND:20250404T095500Z
DTSTAMP:20260421T120133Z
UID:session/Sbr46b6VopVnVZL5jw4xjk@hasgeek.com
SEQUENCE:9
CATEGORIES:Small language models,Hands-on workshop ,Blr OSAI meetup in Apr
 il 2025
CREATED:20250401T084607Z
DESCRIPTION:Talk Overview\nSince its inception\, Sarvam has been pioneerin
 g work on Speech and Text models in Indian languages\, and now we are lead
 ing the efforts on developing foundational models for India. We have relea
 sed 2 major models: Sarvam-1\, which is a 2B parameter model\, delivering 
 SOTA performance on the top 10 Indian languages and English. It is also ab
 le to give comparable performance to bigger 7B and 8B models\, while requi
 ring lower compute and costs for inference. Details: https://huggingface.c
 o/sarvamai/sarvam-1\n\nAnother model is Shuka\, which is an audio-native l
 anguage model which accepts speech inputs and gives text output. The model
  is able to deliver SOTA performance on 10 Indian languages + English\, an
 d is already in the ranks of Gemini Flash\, Llama3 and GPT-40-mini on Indi
 an speech datasets\, while only being an 8B parameter model. It was recent
 ly featured by Meta AI as a part of its globel GenAI success stories blog:
  https://ai.meta.com/blog/sarvam-india-audio-language-model-llama/\n\nKey 
 takeaways\n1. Spread awareness about the innovation in Indian foundational
  models\n2. Encourage develop community to build on top of these foundatio
 nal models and how Sarvam can support them\n\nTarget Audience\n1. Develope
 rs building applications on top of speech APIs and foundational models\n2.
  Enterprises building for Indian customers and looking for Voice AI soluti
 ons\n\nSpeaker Bio:\nTushar Goswamy: I am working on GTM & Strategy at Sar
 vam\, and leading some of our new verticals like Edge AI. I am also leadin
 g partnerships for Sarvam with leading companies like Meta & Google as wel
 l as with select government departments like Ministry of Skill Development
 . I was previously working as a Sr. Consultant with McKinsey\, where I led
  new business development projects in sectors like Electric Vehicles\, Dro
 nes & GenAI. I did my undergrad from IIT Kanpur\, where I co-founded 2 AI 
 startups and dabbled with AI research\, having published at conferences li
 ke NeurIPS\, ICML and COLING. 
GEO:12.9418972;77.6941002
LAST-MODIFIED:20250401T085557Z
LOCATION:9th Floor\, MERCURY BLOCK - Nutanix Technologies India Pvt Ltd\nB
 engaluru\nIN
ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com
URL:https://hasgeek.com/fifthelephant/osai-blr-meetup/schedule/genai-for-a
 -billion-indians-building-text-and-voice-based-llms-for-bharat-Sbr46b6VopV
 nVZL5jw4xjk
BEGIN:VALARM
ACTION:display
DESCRIPTION:Building Bharat’s LLMs: speech & text for a billion in 9th F
 loor\, MERCURY BLOCK in 5 minutes
TRIGGER:-PT5M
END:VALARM
END:VEVENT
BEGIN:VEVENT
SUMMARY:Challenges in building multi-platform on device AI applications
DTSTART:20250404T095500Z
DTEND:20250404T100500Z
DTSTAMP:20260421T120133Z
UID:session/UWhKZfJjJXAYVLc1F696mR@hasgeek.com
SEQUENCE:6
CATEGORIES:AI and costs - optimization\, efficiency\, learnings,10 mins Li
 ghtning talk/demo,Blr OSAI meetup in April 2025
CREATED:20250401T084645Z
DESCRIPTION:## Description\nDeveloping cross-platform AI applications pres
 ents unique challenges\, especially when aiming to provide consistent expe
 riences while leveraging native capabilities. This session explores the ar
 chitectural decisions behind Bodhi App\, an open-source platform that enab
 les users to run LLMs locally. I'll share our experience using the Tauri f
 ramework—combining Rust's performance with web technologies—to create 
 a foundation for multi-platform support while currently focusing on macOS.
 \n\nI'll demonstrate how we structured our application to separate platfor
 m-specific code from core functionality\, enabling future expansion to Win
 dows and Linux\, and possibly Android and iOS. The presentation will cover
  our integration with llama.cpp\, explaining how we handle the communicati
 on between our Rust backend and the inference engine\, while using React a
 nd modern web technologies for the user interface. This architecture allow
 s us to deliver a lightweight application that utilizes native webviews ra
 ther than bundling entire browser engines\, resulting in better performanc
 e and smaller distribution packages.\n\n## Takeaways\n1. Practical insight
 s into using Tauri for cross-platform development\, including how to struc
 ture applications for future platform expansion while initially targeting 
 a single platform.\n\n2. Strategies for integrating AI inference engines l
 ike llama.cpp with modern application architectures that balance native pe
 rformance with cross-platform compatibility.\n\n## Target Audience\nThis s
 ession will benefit developers working on cross-platform applications\, pa
 rticularly those interested in AI integration\, Rust-based architectures\,
  or alternatives to Electron. Software engineers looking to understand how
  to leverage web technologies without sacrificing native performance will 
 find practical approaches they can apply to their own projects.\n\n## Bio\
 nAmir is the founder of Bodhi App\, an open-source platform for local LLM 
 inference. With over 18 years of experience in software architecture\, he 
 specializes in developing applications that bridge high-performance backen
 d systems with accessible user interfaces. His current work focuses on mak
 ing advanced AI technologies available to everyone while preserving privac
 y and eliminating ongoing API costs.\n
GEO:12.9418972;77.6941002
LAST-MODIFIED:20250401T130113Z
LOCATION:9th Floor\, MERCURY BLOCK - Nutanix Technologies India Pvt Ltd\nB
 engaluru\nIN
ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com
URL:https://hasgeek.com/fifthelephant/osai-blr-meetup/schedule/challenges-
 in-building-multi-platform-on-device-ai-applications-UWhKZfJjJXAYVLc1F696m
 R
BEGIN:VALARM
ACTION:display
DESCRIPTION:Challenges in building multi-platform on device AI application
 s in 9th Floor\, MERCURY BLOCK in 5 minutes
TRIGGER:-PT5M
END:VALARM
END:VEVENT
BEGIN:VEVENT
SUMMARY:Q&A for demos
DTSTART:20250404T100500Z
DTEND:20250404T101000Z
DTSTAMP:20260421T120133Z
UID:session/AqA8DnDU4jyNPAPvQXXevV@hasgeek.com
SEQUENCE:3
CREATED:20250401T084703Z
GEO:12.9418972;77.6941002
LAST-MODIFIED:20250401T085133Z
LOCATION:9th Floor\, MERCURY BLOCK - Nutanix Technologies India Pvt Ltd\nB
 engaluru\nIN
ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com
BEGIN:VALARM
ACTION:display
DESCRIPTION:Q&A for demos in 9th Floor\, MERCURY BLOCK in 5 minutes
TRIGGER:-PT5M
END:VALARM
END:VEVENT
BEGIN:VEVENT
SUMMARY:Tea break
DTSTART:20250404T101000Z
DTEND:20250404T103000Z
DTSTAMP:20260421T120133Z
UID:session/LfHBCue3ZZJey2M89HUP5G@hasgeek.com
SEQUENCE:5
CREATED:20250401T084714Z
GEO:12.9418972;77.6941002
LAST-MODIFIED:20250401T085140Z
LOCATION:9th Floor\, MERCURY BLOCK - Nutanix Technologies India Pvt Ltd\nB
 engaluru\nIN
ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com
BEGIN:VALARM
ACTION:display
DESCRIPTION:Tea break in 9th Floor\, MERCURY BLOCK in 5 minutes
TRIGGER:-PT5M
END:VALARM
END:VEVENT
BEGIN:VEVENT
SUMMARY:Sponsored talk: Transformers showdown: advances\, trade-offs\, and
  insights
DTSTART:20250404T103000Z
DTEND:20250404T105500Z
DTSTAMP:20260421T120133Z
UID:session/JHz4Qx5Ekh4Q59eATZPuTX@hasgeek.com
SEQUENCE:7
CATEGORIES:Other,Sponsored talk (25-35 mins),Blr OSAI meetup in April 2025
CREATED:20250401T084741Z
DESCRIPTION:Transformers have evolved a lot in the last few years ever sin
 ce the introduction in 2017.  We take a look at such different architectur
 es and dive into what are the benifits and drawbacks of each.\n\n### Agend
 a\n1. Transformer architecture overview\n2. How attention mechanism evolve
 d over time\n     a. MHA\, MQA and GQA\n     b. MLA from DeepSeek\n4. Curr
 ent improvements like Diff Transformer and nGPT\n5. How the differences st
 ack up against each other.\n6. What are the inference time implications of
  each.\n7. Comparison on small scale models.\n8. Analysis and Conclusion\n
 \n### Takeaways\nIntution and Functional explanation as to why changes hav
 e been incorporated since the inception of transformers. \n\n### Target au
 dience\nPeople willing to understand and learn the evolution of transforme
 rs and possible future. ML enthusiasts and professionals.
GEO:12.9418972;77.6941002
LAST-MODIFIED:20250401T130213Z
LOCATION:9th Floor\, MERCURY BLOCK - Nutanix Technologies India Pvt Ltd\nB
 engaluru\nIN
ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com
URL:https://hasgeek.com/fifthelephant/osai-blr-meetup/schedule/transformer
 -architectures-royal-rumble-JHz4Qx5Ekh4Q59eATZPuTX
BEGIN:VALARM
ACTION:display
DESCRIPTION:Sponsored talk: Transformers showdown: advances\, trade-offs\,
  and insights in 9th Floor\, MERCURY BLOCK in 5 minutes
TRIGGER:-PT5M
END:VALARM
END:VEVENT
BEGIN:VEVENT
SUMMARY:Transition
DTSTART:20250404T105500Z
DTEND:20250404T110000Z
DTSTAMP:20260421T120133Z
UID:session/9unTceinfmS1LsAHbSTunh@hasgeek.com
SEQUENCE:2
CREATED:20250401T084818Z
GEO:12.9418972;77.6941002
LAST-MODIFIED:20250401T085145Z
LOCATION:9th Floor\, MERCURY BLOCK - Nutanix Technologies India Pvt Ltd\nB
 engaluru\nIN
ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com
BEGIN:VALARM
ACTION:display
DESCRIPTION:Transition in 9th Floor\, MERCURY BLOCK in 5 minutes
TRIGGER:-PT5M
END:VALARM
END:VEVENT
BEGIN:VEVENT
SUMMARY:Automating slide analysis with Granite Vision AI
DTSTART:20250404T110000Z
DTEND:20250404T111000Z
DTSTAMP:20260421T120133Z
UID:session/38RnTZSKKuLMDuGd6P9EY5@hasgeek.com
SEQUENCE:8
CATEGORIES:Real-life uses of problems AI tech solves,Sponsored pitch (10 m
 ins),Blr OSAI meetup in April 2025
CREATED:20250401T084808Z
DESCRIPTION:## Abstract\nThe Granite Vision model\, part of IBM's Granite 
 family of open-source foundation models\, enables state-of-the-art image a
 nalysis and understanding. With over 2 billion parameters\, the model is d
 esigned for high-performance computer vision tasks\, including object dete
 ction\, scene understanding\, and visual question answering. This session 
 demonstrates how to harness the power of the Granite Vision model to build
  a PowerPoint (PPT) AI image analysis and question-answering system. By co
 mbining advanced vision capabilities with natural language processing\, th
 is system automates insights extraction from presentation slides\, offerin
 g immense value to startups\, enterprises\, and educators. \n\nFor the lea
 dership perspective\, this solution provides tangible benefits\, such as a
 utomating repetitive slide review tasks\, saving valuable time\, and enabl
 ing faster decision-making. By leveraging cutting-edge AI for actionable i
 nsights from unstructured visual data\, startups can drive innovation and 
 gain a competitive edge in their industries.\n\n\n### Concepts for Buildin
 g the PPT AI Analyzer\n- Leverage IBM’s Granite Vision model with 2 bill
 ion parameters for advanced image analysis tasks.\n- Extract and classify 
 visual elements like charts\, graphs\, and tables from PowerPoint slides.\
 n- Integrate AI-driven image recognition with natural language processing 
 for question answering.\n- Automate insights extraction from presentation 
 slides to streamline workflows.\n\n## Takeaways\n- Learn how the Granite V
 ision model revolutionizes image analysis and question answering for enter
 prise use cases.\n- Understand the architecture and practical implementati
 on of a scalable AI-powered system for analyzing PowerPoint presentations.
 \n- Gain insights into real-world applications and how entrepreneurs can a
 dopt this solution to drive innovation and efficiency.\n\n## Which Audienc
 es is Your Session Going to Benefit?\n- Organization Leadership: Gain stra
 tegic insights into leveraging AI for business innovation and efficiency.\
 n- AI Engineers and Developers: Understand the technical architecture and 
 implementation details.\n- Educators and Analysts: Explore automated tools
  for interactive content creation and report analysis.\n\n*****\n\n## Addi
 tional Resources\n- **Granite Vision Model on Hugging Face:**  Explore IBM
 ’s state-of-the-art Granite Vision model with 2 billion parameters\, des
 igned for advanced image analysis and multimodal tasks. Explore [Granite V
 ision Model on Hugging Face][granite_vision_link]\n\n[granite_vision_link]
 : https://huggingface.co/ibm-granite/granite-vision-3.2-2b\n\n- **IBM Tuto
 rial: PPT AI Analyzer:** Step-by-step guide to building a PowerPoint AI An
 alyzer with the Granite Vision model. Walk through the full [tutorial - Bu
 ild a PPT AI Analyzer][tutorial_link]\n\n[tutorial_link]: https://www.ibm.
 com/think/tutorials/ppt-ai-analyzer\n\n*****\n## Speaker's Bio\nVrunda Gad
 esha - AI Adovate | IBM \n\nShe is a Data Scientist\, Ph.D. scholar\, and 
 AI enthusiast with expertise in Large Language Models\, Natural Language P
 rocessing\, Machine Learning\, and technical content creation. Skilled in 
 Python Programming\, she has led AI solution development and shared her kn
 owledge through academic writing and corporate training. She is passionate
  about advancing AI and data science and is committed to continuous learni
 ng and impactful innovation.
GEO:12.9418972;77.6941002
LAST-MODIFIED:20250401T130322Z
LOCATION:9th Floor\, MERCURY BLOCK - Nutanix Technologies India Pvt Ltd\nB
 engaluru\nIN
ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com
URL:https://hasgeek.com/fifthelephant/osai-blr-meetup/schedule/build-a-ppt
 -ai-image-analysis-question-answering-system-with-granite-vision-model-38R
 nTZSKKuLMDuGd6P9EY5
BEGIN:VALARM
ACTION:display
DESCRIPTION:Automating slide analysis with Granite Vision AI in 9th Floor\
 , MERCURY BLOCK in 5 minutes
TRIGGER:-PT5M
END:VALARM
END:VEVENT
BEGIN:VEVENT
SUMMARY:Streamlining LLM data pipelines with data prep kit
DTSTART:20250404T111000Z
DTEND:20250404T112000Z
DTSTAMP:20260421T120133Z
UID:session/FQLbYTCn82AmvcASyjda7n@hasgeek.com
SEQUENCE:6
CATEGORIES:Data collation\, data cleaning,Sponsored talk (25-35 mins),Blr 
 OSAI meetup in April 2025
CREATED:20250401T084844Z
DESCRIPTION:## Abstract\nPreparing high-quality datasets is a critical yet
  time-consuming process when building large language models (LLMs). Data P
 rep Kit\, an open-source Python toolkit\, simplifies and automates data pr
 eparation tasks\, enabling faster and more efficient workflows for LLM app
 lications. This session will explore how Data Prep Kit addresses key chall
 enges like text extraction\, deduplication\, and data quality scoring\, al
 ong with insights from real-world use cases such as creating an exclusive 
 RAG. Attendees will learn how to leverage the toolkit to streamline their 
 data pipelines\, enhance dataset quality\, and maximize the efficiency of 
 LLM development.\n\n### Overview of Features and Concepts for Data Prepara
 tion\n- **Text Extraction from Diverse Formats** - Extracts text from PDFs
 \, HTML\, Word documents\, and other formats while preserving document str
 ucture like tables and headings.\n- **Duplicate and Near-Duplicate Removal
 ** - Uses hash-based exact deduplication and fuzzy deduplication with MinH
 ash to eliminate redundant data and improve dataset quality.\n- **Document
  Quality Scoring** - Calculates metrics like word count\, bad word detecti
 on\, placeholder text (e.g.\, Lorem Ipsum)\, and bullet point ratios to fi
 lter out low-quality data.\n- **PII and Malicious Content Removal** - Iden
 tifies and removes personal identifiable information (PII) and malicious c
 ode to ensure datasets are safe and compliant.\n- **Scalability for Large 
 Datasets** - Runs efficiently on laptops for small datasets and scales sea
 mlessly to distributed environments like Ray and Spark for large-scale dat
 a preparation.\n- **Integration with Open-Source Tools** - Leverages libra
 ries like DocLing for text extraction and ClamAV for malware detection\, m
 aking it extensible and modular.\n- **Real-World Use Case** - Successfully
  used to prepare terabytes of data for IBM’s Granite LLM\, demonstrating
  its effectiveness in enterprise-scale applications.\n\n## Takeaways\n- A 
 comprehensive understanding of how to use Data Prep Kit to automate and en
 hance data preparation for LLM pipelines.\n- Insights into best practices 
 for cleaning\, deduplicating\, and scoring data quality.\n- Practical know
 ledge of scalable workflows for preparing multimodal datasets.\n\n## Which
  Audiences is Your Session Going to Benefit?\n- AI engineers and practitio
 ners working on training or fine-tuning LLMs.\n- Data scientists and data 
 engineers responsible for preparing datasets for machine learning applicat
 ions.\n\n****\n\n## Additional Resources\nHere are some key resources to e
 xplore and understand the capabilities of *Data Prep Kit* and its applicat
 ions in LLM development:\n- **GitHub Repository:** - Access the full sourc
 e code\, documentation\, and examples for Data Prep Kit on GitHub. Check t
 his link to access [Data Prep Kit GitHub Repository.][github_repo_link]\n 
  \n[github_repo_link]: https://github.com/data-prep-kit/data-prep-kit\n\n-
  **IBM Granite Open Source LLM:** - The transforms developed in this toolk
 it were instrumental in preparing data for IBM’s Granite LLM models\, no
 w available on Hugging Face. Click here to access [IBM Granite Models from
  HuggingFace.][granite_model_link]\n\n[granite_model_link]: https://huggin
 gface.co/ibm-granite\n\n- **LF AI & Data Foundation:** - Data Prep Kit is 
 hosted as a project under the LF AI & Data Foundation\, reflecting its imp
 ortance and adoption in the open-source AI community. Click here to check 
 out [LF AI & Data Foundation][foundation_link]\n\n[foundation_link]: https
 ://lfaidata.foundation/projects/\n\n- **IBM Developer Learning Path:** - A
  step-by-step guide to get started with Data Prep Kit\, including details 
 on its architecture and sample use cases like RAG and fine-tuning with rea
 l-world data. Checkout the [IBM Developer learning path][learning_path_lin
 k]\n\n[learning_path_link]: https://developer.ibm.com/learningpaths/get-st
 arted-data-prep-kit/\n\nThese resources provide technical depth\, practica
 l examples\, and community-driven insights to help you fully leverage Data
  Prep Kit in your projects.\n\n## Speaker's Bio\nVrunda Gadesha - AI Advoc
 ate | IBM \n\nShe is a Data Scientist\, Ph.D. scholar\, and AI enthusiast 
 with expertise in Large Language Models\, Natural Language Processing\, Ma
 chine Learning\, and technical content creation. Skilled in Python Program
 ming\, she has led AI solution development and shared her knowledge throug
 h academic writing and corporate training. She is passionate about advanci
 ng AI and data science and is committed to continuous learning and impactf
 ul innovation.\n\n#### *Note: We are open for the full 30 mins talk as wel
 l as a lightning talk as well for this topic.*\n
GEO:12.9418972;77.6941002
LAST-MODIFIED:20250401T130331Z
LOCATION:9th Floor\, MERCURY BLOCK - Nutanix Technologies India Pvt Ltd\nB
 engaluru\nIN
ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com
URL:https://hasgeek.com/fifthelephant/osai-blr-meetup/schedule/preparing-d
 ata-for-llm-applications-using-data-prep-kit-FQLbYTCn82AmvcASyjda7n
BEGIN:VALARM
ACTION:display
DESCRIPTION:Streamlining LLM data pipelines with data prep kit in 9th Floo
 r\, MERCURY BLOCK in 5 minutes
TRIGGER:-PT5M
END:VALARM
END:VEVENT
BEGIN:VEVENT
SUMMARY:Q&A for demos
DTSTART:20250404T112000Z
DTEND:20250404T112500Z
DTSTAMP:20260421T120133Z
UID:session/FB9wKgFn2b2njDNJauedy7@hasgeek.com
SEQUENCE:4
CREATED:20250401T084859Z
GEO:12.9418972;77.6941002
LAST-MODIFIED:20250401T085154Z
LOCATION:9th Floor\, MERCURY BLOCK - Nutanix Technologies India Pvt Ltd\nB
 engaluru\nIN
ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com
BEGIN:VALARM
ACTION:display
DESCRIPTION:Q&A for demos in 9th Floor\, MERCURY BLOCK in 5 minutes
TRIGGER:-PT5M
END:VALARM
END:VEVENT
BEGIN:VEVENT
SUMMARY:Break
DTSTART:20250404T112500Z
DTEND:20250404T114000Z
DTSTAMP:20260421T120133Z
UID:session/Xfb91w9Riu35KFZ3M1P7u3@hasgeek.com
SEQUENCE:3
CREATED:20250401T085106Z
GEO:12.9418972;77.6941002
LAST-MODIFIED:20250401T085200Z
LOCATION:9th Floor\, MERCURY BLOCK - Nutanix Technologies India Pvt Ltd\nB
 engaluru\nIN
ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com
BEGIN:VALARM
ACTION:display
DESCRIPTION:Break in 9th Floor\, MERCURY BLOCK in 5 minutes
TRIGGER:-PT5M
END:VALARM
END:VEVENT
BEGIN:VEVENT
SUMMARY:Smarter test automation with AI agents & Hercules
DTSTART:20250404T114000Z
DTEND:20250404T115000Z
DTSTAMP:20260421T120133Z
UID:session/ESYr6kPfNYu3ji9gu8WbWB@hasgeek.com
SEQUENCE:5
CATEGORIES:Next generation architectures,Speaking at the Fifth Elephant 20
 25 Annual Conference,30 mins talk
CREATED:20250401T085223Z
DESCRIPTION:## 🔍 About the review  \n\nModern test automation tools fac
 e two key challenges 🧩:  \n\n-  **Complexity:** Non-technical teams str
 uggle with existing tools.  \n-  **Maintenance:** Traditional test executi
 on lacks intelligent state based execution and interactions.  \n\nThese in
 efficiencies delay releases\, increase costs\, and reduce software quality
 .  \n\n**Hercules**\, an open-source project by TestZeus\, redefines end-t
 o-end testing by leveraging AI agents to:  \n✅ Execute natural language 
 tests\n✅ Interact with browsers\, APIs\, or other interfaces  \n✅ Bid 
 adieu to automation maintenance\, as we see how multi agentic systems "REA
 ct" to the application at hand.\n\nThis presentation will showcase how Her
 cules tackles core testing challenges\, its value to the software developm
 ent lifecycle\, and its impact on the modern testing landscape.  \n\n## 
 🚀 About Hercules: world's first open source end to end testing agent  \
 n\nHercules is the world’s first open-source testing agent\, built to ha
 ndle the toughest testing tasks so you don’t have to. It turns simple\, 
 easy-to-write Gherkin steps into fully automated end to end tests—no cod
 ing skills needed. Whether you’re working with tricky platforms like Sal
 esforce or running tests in your CI/CD pipeline\, Hercules adapts to your 
 needs and takes care of the details. With Hercules\, testers can perform U
 I/API/Accessibility/Security/Mobile and **Visual testing**. All of this th
 anks to a multi agentic framework under the hood. \nIn this session \, we 
 will pop the hood\, and see how this multiagentic solution works by utiliz
 ing a small army of agents like Browser agent\, API agent\, etc. and tools
  like Playwright to ensure that the user can tackle the most herculean tas
 ks of test automation. \n\n\n## 🎯 Key takeaways for the audience  \n- W
 arm introduction to multi-agentic architectures.\n- Notes on cost and prom
 pt optimization\n- Deploying agents to production and "gotchas" \n- Buildi
 ng in Open Source and Building for Extensibility.  \n\n## 🎯 Target audi
 ence  \n- Testers  \n- Developers \n- QA Managers  \n- AI/ML Practitioners
   \n- Open-source Enthusiasts  \n\n## 📌 GitHub Repository  \n🔗 [Herc
 ules on GitHub](https://github.com/test-zeus-ai/testzeus-hercules/tree/mai
 n)  \n\n## 🎤 About the Presenter  \n\n[Robin Gupta](https://www.robin-g
 upta.com) is the human at the helm for TestZeus\, the company behind world
 's first open source testing agent. He is a versatile engineering leader w
 ith more than 15 years of experience in software delivery across startups\
 , scale-ups and enterprises. He is a System 1 thinker\, and has led produc
 ts contributing to $10M ARR. With a metrics-driven approach\, he has eleva
 ted engineering maturity of product teams for diverse domains such as BFSI
 \, EdTech\, Retail\, and Developer Experience. Beyond work\, he mentors at
  ADPList and Plato\, contributes to open-source projects like Selenium and
  has authored books and international courses on software testing. He is a
 lso a recognized speaker at international events such as Dreamforce (by Sa
 lesforce) and Selenium Conference.
GEO:12.9418972;77.6941002
LAST-MODIFIED:20250401T130406Z
LOCATION:9th Floor\, MERCURY BLOCK - Nutanix Technologies India Pvt Ltd\nB
 engaluru\nIN
ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com
URL:https://hasgeek.com/fifthelephant/osai-blr-meetup/schedule/worlds-firs
 t-open-source-testing-agent-ESYr6kPfNYu3ji9gu8WbWB
BEGIN:VALARM
ACTION:display
DESCRIPTION:Smarter test automation with AI agents & Hercules in 9th Floor
 \, MERCURY BLOCK in 5 minutes
TRIGGER:-PT5M
END:VALARM
END:VEVENT
BEGIN:VEVENT
SUMMARY:Simplifying legal help with AI-powered law matching
DTSTART:20250404T115000Z
DTEND:20250404T120000Z
DTSTAMP:20260421T120133Z
UID:session/5WyB5ByiVwvyzXcAnHYDaR@hasgeek.com
SEQUENCE:3
CATEGORIES:Finetuning LLMs,Birds of Feather (BOF) session ,Blr OSAI meetup
  in April 2025
CREATED:20250401T085237Z
DESCRIPTION:{LLMs are well equipped to solve tasks that require world know
 ledge. There is a considerable gap between frontier LLMs on English langua
 ge vs the best language models we have in Indic languages. While there is 
 considerable ongoing effort towards merging this gap\, a more practical an
 d scalable solution approach comes through a well designed alignment: wher
 e an existing LLM is stratetically tuned for optimal performance on a doma
 in and/or task in indic language. InstructLAB (https://github.com/instruct
 lab) is an open source tool which is designed to make alignment very easy 
 for developers\, data scientists and any AI practioner at SME roles withou
 t requiring AI/ML technical skills. It is a developed as community project
  soliciting contributions from practioners solving real world problems via
  LLM alignment. In this work we will demonstrate how InstructLAB can be us
 ed to solve one such real world problem: Legal Statue Identification in In
 dian context from a common Indian citizen's perspective and in both Englis
 h and Hindi. This work is co-developed by IBM research\, IITB\, BharatGen.
  \n\n\nLegal Statue Identification (LSI) is a known task in legal domain w
 here summaries from court proceedings is used to identify what all relevan
 t laws gets applicable for the proceedings. Summaries are often full of le
 galese as prepared by dedicated personnel attending court proceedings and 
 thus it requires legal experts to figure out relevant laws being applicabl
 e across various segments of the summary. In contrast\, various legal help
  websites such as "Indian Kanoon" or "LawRato"\, "LegalKart" are thriving 
 with examples of posts coming from Indian citizens describing their proble
 m and seeking legal advice as per Indian law from designated legal experts
 . In this session we will show how we can use "InstructLAB"\, an open sour
 ce language model alignment tool\, to align a small embedding model toward
 s this specific use-case : Given a complaint description in ordinary citiz
 en's langauage\,  retrieve relevant laws from Bharatiya Nyay Sanhita (BNS)
  as applicable for the complaint scenario. More specifically\, we will sho
 w how InstructLAB can be used to generate diverse and strategically curate
 d synthetic data over the entire BNS corpus and how good quality synthetic
 ally generated data is enough to tune a small embedding model of only ~560
 M to accurately do law retrieval and thus solving LSI task for common citi
 zens.   your session in 2 paragraphs}\n\n{1. InstrcutLAB is an efficient a
 nd easy to use tool for domain alignment.\n2. Legal Statue Identification 
 starting from Citizen's grievance can be solved by aligning a small embedd
 ing model over India's latest legal document BNS(Bharatiya Nyay Sanhita). 
 \n}\n\n{The major audience would be AI practioners including developers\, 
 SMEs\, CTOs looking to use LLMs on specific domains/tasks. Also working pr
 ofessionals in legal domain will find this application useful. }\n\n{I'm J
 aydeep Sen\, a senior research scientist at IBM Research\, India and manag
 er of speech and NLP technologies team there. Jaydeep primarily works on Q
 uestion Answering and related NLP technologies and is currently engaged in
  cutting edge research on designing robust neural retrievers. He has autho
 red 20+ papers in top conferences like ACL\, EMNLP\, IJCAI\, VLDB\, SIGMOD
  etc and 30+ patents and a designated Master Inventor at IBM.\n\nCollabora
 tors: \nIBM Research: Rudra Murthy\, Ashish Mittal\, Sachindra Joshi\, Ami
 th SInghee\nBharatGen: Amrith Krishna\, Kundeshwar Pundalik\, Ajay Nagpal\
 , Piyushi Anand\, Smitha Gautam\, John Nirmal \nIITB: Ganesh Ramakrishnan\
 n }
GEO:12.9418972;77.6941002
LAST-MODIFIED:20250401T085801Z
LOCATION:9th Floor\, MERCURY BLOCK - Nutanix Technologies India Pvt Ltd\nB
 engaluru\nIN
ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com
URL:https://hasgeek.com/fifthelephant/osai-blr-meetup/schedule/know-your-l
 aws-in-bharatiya-nyay-sanhita-5WyB5ByiVwvyzXcAnHYDaR
BEGIN:VALARM
ACTION:display
DESCRIPTION:Simplifying legal help with AI-powered law matching in 9th Flo
 or\, MERCURY BLOCK in 5 minutes
TRIGGER:-PT5M
END:VALARM
END:VEVENT
BEGIN:VEVENT
SUMMARY:Q&A for demos
DTSTART:20250404T120000Z
DTEND:20250404T120500Z
DTSTAMP:20260421T120133Z
UID:session/AUjULpqzwhPPDbTX5P7ejN@hasgeek.com
SEQUENCE:0
CREATED:20250401T085250Z
GEO:12.9418972;77.6941002
LAST-MODIFIED:20250401T085250Z
LOCATION:9th Floor\, MERCURY BLOCK - Nutanix Technologies India Pvt Ltd\nB
 engaluru\nIN
ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com
BEGIN:VALARM
ACTION:display
DESCRIPTION:Q&A for demos in 9th Floor\, MERCURY BLOCK in 5 minutes
TRIGGER:-PT5M
END:VALARM
END:VEVENT
BEGIN:VEVENT
SUMMARY:BOF: Multi agent systems - the future of computer vision
DTSTART:20250404T120500Z
DTEND:20250404T123500Z
DTSTAMP:20260421T120133Z
UID:session/VjKSFq7oH6V5WHNPg3ifEG@hasgeek.com
SEQUENCE:4
CATEGORIES:Agentic AI,Sponsored workshop,Blr OSAI meetup in April 2025
CREATED:20250401T085258Z
DESCRIPTION:\n\nThe core advantage of multi-agent systems in computer visi
 on lies in their ability to divide complex tasks into smaller\, manageable
  sub-tasks\, allowing for more efficient and scalable processing. This par
 adigm fosters collaboration\, where agents can exchange information and up
 date each other’s knowledge\, resulting in a more refined outcome\, keep
 ing every agrnt in sync.\n\nFor instance\, in generative tasks\, one agent
  might focus on generating the background while another agent handles fore
 ground details or texture. The key to success in these systems is their ab
 ility to dynamically coordinate actions\, allowing the agents to adapt to 
 new visual contexts\, learn from each other\, and provide creative\, high-
 quality results in real-time. Libraries like smolagents\, crewai have prov
 ided platforms to build and experiment with multi-agent setups\n\nKey Take
 aways:\n1. Libraries and Frameworks: Popular libraries like Ray smolagents
 \, browseruse\, CrewAI enable the development and experimentation with MAS
  in computer vision applications (will discuss these frameworks in detail)
 .\n\n2. Experimentation and Innovation: Experiments in MAS-based computer 
 vision have led to breakthroughs in areas like real-time image generation 
 and adaptive learning\, where agents learn to collaborate and improve over
  time\, offering more dynamic and context-aware outputs.\n\n3. How to buil
 d robust and scalable systems around MAS (Multi Agentic Systems) to scale 
 to millions of users\n\n4. How to detect and handle failures/hallucination
 s in thise systems\n\nI am Akshat. I work as Tech Lead\, Machine Learning 
 at Glance. InMobi\, working on a few generative AI tracks involving vision
 \, audio and RL.\nhttps://www.linkedin.com/in/agupta28/
GEO:12.9418972;77.6941002
LAST-MODIFIED:20250401T085825Z
LOCATION:9th Floor\, MERCURY BLOCK - Nutanix Technologies India Pvt Ltd\nB
 engaluru\nIN
ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com
URL:https://hasgeek.com/fifthelephant/osai-blr-meetup/schedule/the-emergen
 ce-of-multi-agent-systems-in-computer-vision-a-new-era-of-creative-collabo
 ration-VjKSFq7oH6V5WHNPg3ifEG
BEGIN:VALARM
ACTION:display
DESCRIPTION:BOF: Multi agent systems - the future of computer vision in 9t
 h Floor\, MERCURY BLOCK in 5 minutes
TRIGGER:-PT5M
END:VALARM
END:VEVENT
BEGIN:VEVENT
SUMMARY:Summary\; thank-yous
DTSTART:20250404T123500Z
DTEND:20250404T124000Z
DTSTAMP:20260421T120133Z
UID:session/GTymfwcLHXobDBWZxCwZjV@hasgeek.com
SEQUENCE:0
CREATED:20250401T085329Z
GEO:12.9418972;77.6941002
LAST-MODIFIED:20250401T085329Z
LOCATION:9th Floor\, MERCURY BLOCK - Nutanix Technologies India Pvt Ltd\nB
 engaluru\nIN
ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com
BEGIN:VALARM
ACTION:display
DESCRIPTION:Summary\; thank-yous in 9th Floor\, MERCURY BLOCK in 5 minutes
TRIGGER:-PT5M
END:VALARM
END:VEVENT
END:VCALENDAR