See the current status of contributions for each language. The most top-3 contributions are for Malayalam, Telugu, and Bengali languages.

My latest for MIT Technology Review What does it take to build a desi ChatGPT, without compute, clean data or even the internet in its own languages? From solo founders training models on fumes to govt-backed missions deploying 19,000 GPUs--this is India’s ambitious, messy quest to build

EKA-Eval : A Comprehensive Evaluation Framework for Large Language Models in Indian Languages First paper and open source contribution out from Project EKA. EKA-Eval is our consolidated evaluation pipeline for LLMs for Indic and non-Indic languages. Covers 35+ benchmarks across

This is officially one of my favourite blog posts. 1 sem of HPC packed into 1 blog github.com/Dao-AILab/quac…

Tokenization has been the final barrier to truly end-to-end language models. We developed the H-Net: a hierarchical network that replaces tokenization with a dynamic chunking process directly inside the model, automatically discovering and operating over meaningful units of data

fwiw the amount of compute needed for Kimi K2 is around about the same as DeepSeek V3/R1 ($5m worth) More stable training, similar active parameters, tokens trained on But much better performance Compute isn’t all you need..

Of course this was expected. One of the strongest models from the West is now moving behind closed doors. No surprise there. On the other side, China is producing some technically solid models—but trusting them for critical or sensitive work? That’s a serious risk most would

When your brain is more active at 2 AM than 8 AM: “2AM Hustleosis” – a chronic compulsion to build at 2 AM and crash at 8. Symptoms include caffeine dependency, late-night genius, and morning regret. #startuplife #founderproblems

Glad to see Invest India and its CEO Nivruti pushing for investments into the Indian AI ecosystem. It’s much needed especially for the kind of R&D efforts we need to build. I can see the momentum with many US, middle east and Singapore VCs reaching out to us. India has all

Maybe India should reachout to host NeurlIPS next year. It will be worth the investment. What do you say IndiaAI?

In the latest announcements at Google I/O Connect 2025, Manish Gupta , Senior Director of Research at Google DeepMind , spotlighted India’s rising momentum in the AI revolution. Showcasing how Google is actively fueling the IndiaAI Mission, he unveiled a series of powerful

OpenAI OpenAI just released two models (120b and 20b MoE) under Apache 2.0 license. First open source post GPT-2. Glad to see open source winning the LLM game 💪🏼 huggingface.co/openai/gpt-oss…

Is there any startup who has forked #KoboToolbox and added #Aadhaar auth into the collection app. Have urgent requirement for a state wide #survey tool. Onboarding within a day. They will have to deploy their solution in the state data center. Budget is decent. Any

OpenAI and ElevenLabs are being so bullish on India, and yet Indian VCs are almost indifferent to supporting research and innovation for frontier models. Need more technical investors please.

The west realising India’s potential. We should have invested in our model 2 years back. We needed it today.

Seems like we are opening doors for China. Good idea to work together on deeptech innovations?

#PaperAcceptance 📢 “COMI-LINGUA: Expert Annotated Large-Scale Dataset for Multitask NLP in Hindi-English Code-Mixing” has been accepted at EMNLP 2025! 📝✨ 🔗 arxiv.org/abs/2503.21670 #NLP #CodeSwitching #EMNLP25 Congrats to all the authors! ✨

So not a tech team. Looks like a govt relations office. OpenAI has been very active with its engagement with GoI.