Open weights, real stakes

Meet-up overview

Recent requests by the Indian government to avoid gold imports highlight how vulnerable India is to its dependence on U.S. dollars for imports. We already depend on other countries for oil and gold. Can we afford to also increase our dependence on foreign models? NVIDIA recently announced RTX Spark, a one-of-a-kind reinvention of the Windows laptop that supports 120B parameter local model inference. Is the future local?

To answer some of these questions, we can look at the experiences of practitioners who have shipped with Open Weights and/or local models.

📅 Schedule is live! Check out the full lineup of talks, timings, and breaks → View schedule

Featured talks

Kanav Dwevedi (The FlyWheel) on how far can you get without fine-tuning: quality as architecture
This talk is a deep dive on building architecture and guardrails around self-hosted open-weight models (Gemma 4) to deliver quality. The FlyWheel runs a production LLM helpline for Gujarati dairy farmers (phone-line and chat) answering questions on animal health, breeding, milk, and schemes, on open-weight models we host ourselves.

Prathamesh Kalamkar (Thomson Reuters) on hard-earned lessons from training domain-specific LLMs: from vocabulary extension to efficient inference
Building high-performing domain-specific LLMs involves far more than picking a foundation model and running supervised fine-tuning — the real challenges lie in data creation, evaluation, continued pretraining, tokenizer design, and inference optimisation. Prathamesh shares practical lessons from multiple LLM projects, including building Aalap (an Indian legal assistant) and extending language models for chemistry applications.

Nilesh Kumar (NVIDIA) on open weights in production: a practitioner’s inference playbook
Running an open-weight model on a laptop is easy; getting it to serve reliably in production - on-prem or on-cloud GPUs - is where most teams stall. This is a hands-on playbook: model selection, quantization, choosing a serving stack (vLLM, TensorRT-LLM), batching and KV-cache behaviour, the latency-vs-throughput trade-off, evals, and routing — using an open-weight model (Nemotron) as the worked example, with real numbers.

Lightning Talks

Product demo talks from engineers at Red Hat

From LLM to SLM: enterprise fine-tuning that keeps open models improving in Production
Sridhar Pillai

Sovereign LLMs already care; NeMo makes it certain: LLM safety on OpenDataHub/OpenShift AI
Abhijit Roy

Sovereign AI at scale: engineering a secure, private inference platform
Ritesh Shah

Addressing the elephant in the room: feature engineering
Aniket Paluskar & Chaitanya Patel

Who should attend

• Engineers and developers building with open weight models
• Data scientists who’ve fine-tuned or evaluated open weight LLMs
• Technical founders shipping AI products outside the big labs API ecosystem
• Practitioners working on Indian language models or regional use cases
• Anyone tired of vendor lock-in and looking for real alternatives

Meet the editorial team

Craig - AI Automations Engineer, Freelance
Kiran - Founder, Unravel.tech
Sitaram - Site Reliability Engineer, NVIDIA

How to attend

This is a free meet-up. Hit register to attend.

Thank you to our venue hosts

Thank you to Sarang Kulkarni from The Fifth Elephant Community and Darshana Radhakrishnan for supporting the meet-up by hosting it at Thoughtworks Pune. Thank you for helping bring the community together.

Need more info?

☎️ Call: (91) 7676332020
📧 Email: info@hasgeek.com