AI Orchestration System as a Digital Public Good: Tech Stack for the Impact Sector

Governments, multilaterals and social orgs in India & developing worlds are enthusiastic about deploying AI to accelerate impact outcomes in agriculture, public health, poverty alleviation etc. However, mainstream AI models do not account for the linguistic needs, low-resource constraints of global majority nations. Impact orgs often lack the technical capacity and budgets to fine-tune for these, while ensuring privacy and sovereignty considerations. The few successes of AI adoption in the field remains siloed.

This session will demonstrate how Gooey.ai built its open-source AI orchestration and hosting platform (certified as a DPG) to accelerate the deployment, iteration and testing of AI-based solutions for the impact space. The talk will feature how we build 

1. An open, interoperable architecture to push for hot swappability and prevent vendor lock-in (especially relevant for the post-Fable world). 

2. A robust model-eval platform to leverage the constant advances in AI models by benchmarking models for your use case. 

3. Evaluate and choose whichever model is best, fastest, cheapest, safest, or most sovereign for the task. 

4. Highly configurable workflows ranging from all cloud to air gapped One click Deployment on Whatsapp, Telegram, Voice, Web and more

5. A growing library of simple, forkable AI workflows (or AI recipes) for orgs to discover, tweak and apply the successful AI interventions that have worked for their peers to their own problems. 

6. Platform catered to non-coders so development orgs can prototype, measure impact and iterate faster economically 

1. On-ground learnings from AI integration in impact orgs across India, Nairobi, Kenya, US, etc

2. AI workflows and orchestration system as an overlooked enabler of SDG-focused knowledge commons and sovereign tech stack

3. Creating architecture of hot-swappability: The point is not to reject private frontier models; but rather to avoid dependency on any single provider and encourage competition. Enshrine interoperability by design.

4. Hybrid approach to data privacy and compute: Orgs should easily be able to run, fork and test AI solutions in the cloud (cheaply) and quickly deploy them to sovereign data centers if needed.

5. Transparent & Forkable AI recipes for impact-oriented use cases (borrowing from stalwarts like GitHub and View Source on HTML)