Writings on design, technology, stuff I built, and other random shower thoughts.
Shortcomings of AI memory, proposal for a better solution.
The realtime multimodal perception system that set Tavus apart, how it was built, mistakes, learnings.
Enterprise RAG in under <40ms, scalable to thousands of documents, as we take advantage of our conversational setup.
Investigating how a large reasoning LM can interact with a small conversational one to produce deeper conversational agents.