TPU Learning Club 2026 Kolkata
TPU Learning Club Kolkata | Accelerate AI with TPUs, JAX & TorchTPU
Step into high-performance hardware acceleration with TPU Learning Club Kolkata, hosted by Machine Learning Kolkata and supported by Google for Developers.
Join developers, researchers, and machine learning engineers on Saturday, November 7, 2026, for a focused, practical meetup kicking off the Google TPU Learning Clubs campaign in Kolkata.
Learn to build, optimize, and scale deep learning workloads across Google Tensor Processing Units (TPUs) using modern frameworks, including JAX, Keras 3, PyTorch/XLA, and TorchTPU.
What We Will Cover (Key Tech Focus)
- TPU Architecture & XLA Compilation: Understand Matrix Multiplication Units (MXUs), memory topology, and how the XLA compiler optimizes computation graphs.
- JAX & TorchTPU Execution: Leverage functional transforms (
jit,vmap,pmap) in JAX and run native PyTorch models on TPU hardware via TorchTPU. - Gemini 4: Explore Gemini 4's advanced reasoning, coding, and multimodal capabilities through live demonstrations and practical experiments.
- Free-Tier LLM Fine-Tuning: Fine-tune open-weight LLMs on free cloud TPU runtimes (Google Colab and Kaggle) without enterprise hardware costs.
- Hardware Bottleneck Troubleshooting: Profile execution, avoid XLA recompilation loops, and optimize data loaders to eliminate input pipeline starvation.
Who Should Attend and Prerequisites
Target Audience: ML engineers, software developers, AI researchers, and students interested in distributed training, hardware acceleration, and cost-effective model scaling.
Prerequisites:
- Colab & Kaggle Accounts: Exercises execute on free cloud TPU environments, requiring zero local setup.
- Foundational Knowledge: Working knowledge of Python and core machine learning concepts.
- A Laptop: Optional.
Agenda
| Time | Session | Description |
|---|---|---|
| 10:00 AM – 10:30 AM | Registration and Check-in | Check in, get seated, verify Colab/Kaggle TPU runtimes, and network with attendees. |
| 10:30 AM – 10:45 AM | Welcome and Introduction | Opening remarks on the Google AI Community Network initiative and the Kolkata TPU roadmap. |
| 10:45 AM – 11:35 AM | Accelerate AI with TPUs, JAX & TorchTPU | Deep dive into TPU hardware architecture, XLA graph compilation, JAX paradigms, and running PyTorch workloads with TorchTPU. |
| 11:35 AM – 12:20 PM | Build with Gemini 4 | Explore Gemini 4's advanced capabilities through live demonstrations, practical experiments, and real-world use cases. Evaluate its strengths, limitations, and potential applications. |
| 12:20 PM – 01:10 PM | TPUs for the Rest of Us: Fine-Tuning LLMs Without a GPU Budget | Live fine-tuning of an open LLM on free-tier TPUs, GPU vs TPU mental models, debugging XLA recompilation overhead, and curing data pipeline starvation. |
| 01:10 PM – 01:30 PM | Q&A, Wrap-up, and Group Photo | Open floor for technical questions, community recap, and group photo. |
| 01:30 PM – 02:00 PM | Networking Break and Refreshments | Refreshments, discussions with speakers, and community networking. |
Featured Speakers and Instructors
Rishiraj Acharya
GDE (Keras, GCP, AI), Hugging Face Fellow, Applied Scientist II @ Eternal (Zomato)
Topic: Accelerate AI with TPUs, JAX & TorchTPU
Overview: Exploring TPU silicon design, XLA graph optimization, and bridging PyTorch and JAX workflows with TorchTPU.
Sitam Meur
AI Engineer @ Daily Dose of Data Science
Topic: Build with Gemini 4
Overview: Explore Gemini 4's advanced capabilities through live demonstrations, practical experiments, and real-world use cases. Discover its potential and evaluate its strengths and limitations.
Gourav Ghosal
Senior Engineer @ Qubrid AI
Topic: TPUs for the Rest of Us: Fine-Tuning LLMs Without a GPU Budget
Overview: Live fine-tuning of an open LLM on free-tier cloud TPUs, navigating the differences from GPU development, and tackling common traps like XLA recompilation delays and idle accelerators.
Highlights of the Day
- Live Code & Demos: Real-time fine-tuning walkthroughs, custom TPU kernel development, and architecture deep dives on cloud TPU instances.
- AI-Assisted TPU Engineering: Discover how AI assistance can support developers in building and optimizing accelerator-based workloads.
- Cost-Conscious Workflows: Techniques to run compute-intensive AI workflows on zero-cost cloud accelerators.
- Performance Profiling: Practical tips to fix input starvation, diagnose performance bottlenecks, and optimize XLA compilation graphs.
- Gemini 4 Live Showcase: Explore Gemini 4's capabilities through practical demonstrations and real-world applications.
- Community Feedback: Share perspectives and evaluate Gemini 4's strengths, limitations, and developer experience.
- Community & Mentorship: Direct access to Google Developer Experts and local AI practitioners.
- ML Kolkata Black Club: An exclusive chance to be a part of the invite-only Machine Learning Kolkata Black Club — a professional-focused group curated for growth, open technical discussions, and industry leadership.
- Food & Refreshments: Snacks and beverages served during the networking session.
Event Details and How to Join
| Detail | Info |
|---|---|
| Date & Time | Saturday, November 7, 2026 · 10:00 AM – 2:00 PM IST |
| Venue | Indian School of Ethical Hacking – ISOEH |
| Organizer | Machine Learning Kolkata (supported by Google for Developers) |
| Entry | Free (Mandatory prior registration) |
| Hashtags | #TPULearningClubs #MLKolkata #GoogleForDevelopers #Gemini |
This is a focused, limited-capacity technical meetup. Applications are reviewed on a rolling basis, and confirmed attendees will receive an email confirmation with venue directions and preparation instructions.
Guest Speakers
