The AI revolution isn’t just about using tools - it’s about inventing them.
In this talk, we’ll deconstruct how AI is built - tracing breakthroughs from BERT's open source release to physics based models like Genesis - to highlight a repeatable method grounded in open collaboration, modular experimentation, and community driven problem solving. We’ll then dive into 10 actionable project ideas - from training foundational models on domain specific knowledge graphs to multi-modal, multi-physical AI to run robots to controlling drone swarms using multi-agent AI.
No prior AI expertise and no GPUs? That’s an advantage since fresh perspectives and scarcity lead to new architectures. You may rethink causal reasoning for medical diagnosis, run models on small devices or design AI benchmarks tailored to Indian cultural contexts. We’ll close with open-source repos, datasets, and playgrounds to turn these ideas into code.
AI’s next breakthrough won’t come from trillion parameter models - it’ll start from your pull request. Let's begin.