The Open-Source AI Ecosystem of India

Open-source AI is emerging within the broader FOSS ecosystem as its own sect. It is different from the status quo in what is valuable to share (not just code), and in terms of its economics. There is great potential for open source contributors and organisations from India to make important contributions to the global body of work on the subject. Already, there are organisations in India sharing their work freely - source code, data, methodologies, research - that are extremely helpful for the ecosystem. 

I want to talk about it and take it one step further - what can we do to make it even better? While collaboration happens between many Indian organisations, can we support new contributors to sustain themselves?

---

This ecosystem - specifically within India - has the potential to mature into a sustainable one that can support and nurture its contributors.

While India is still GPU-poor, we are directing a lot of money, talent, and effort to niches that won't be useful to the GPU-rich English-speaking countries. It's easy to imagine that after a few iterations of refinement and learning, organisations here will be the pioneers in bringing AI to a data-poor and GPU-poor country.

There's an omen of stability in the ecosystem in governments and large corporations (with deep $$ pockets) funding the efforts. They don't run like VC funds are bring guarantee to the ecosystem.

For example, here are some examples of organisations in this space that have received $$ from large organisations (likely stable sources of funding):

• Karnataka State Govt. directed ₹170Cr ($20M) to ARTPARK, IISc [^1]
• MEITy is maintaining Bhashini initiative to collect Indic language data, train models, and provide easy access. Supported by Ekstep Foundation (Nilekani) [^2]
• AI4Bharat - backed by MSFT, Google, Nvidia, Bhashini, & EkStep - has been an early organisation in the space and their open-source work is being used within pretty much all other orgs working with Indic languages [^3]
• ... and more

The trend among these organisations - across government-run projects, research labs, and the startups they take help from - is to share their work for others to build upon. It need not always be source code that is valuable to share. With the same principles of freedom as FOSS, some share model weights, data[^4], methodologies[^5], research[^6,8], and even compute[^7]. 

There are more players in the ecosystem - startups, contributors from organisations that use (but don't own) an open source project, and to some extent individuals. At this nascent stage, we can shape this budding ecosystem that has a huge volume of talent and a trend towards being open & public. We can address the imperfections and make it better.

For instance, at Gooey.AI[^9], where I work, we host different models and offer them as APIs and web tools from our service. We do a revenue share with model creators where we pay them from the revenue we generate using their model. I want to be loud about it because we have an opportunity to make a sustainable ecosystem built on FOSS principles.

[^1]: https://indianexpress.com/article/cities/bangalore/artificial-intelligence-robotics-technology-park-launched-iisc-bengaluru-7819484/

[^2]: https://analyticsindiamag.com/indias-project-bhashini-breaking-the-language-barrier-with-ai/

[^3]: https://analyticsindiamag.com/why-india-needs-more-ai4bharats/

[^4]: https://ai4bharat.iitm.ac.in/resources/datasets/

[^5]: https://www.karya.in/

[^6]: https://ai4bharat.iitm.ac.in/publications/

[^7]: https://peopleplus.ai/occ

[^8]: https://artpark.in/publications

[^9]: https://gooey.ai