hi i'm prasad
engineering AI products & the systems around them.
Blog
How Prasad Sankar made a production voice AI system respond faster by measuring the full conversation, removing repeated work, and streaming every stage.
How to move sensitive data between two devices with no server in the middle - encrypted at rest, post-quantum safe in transit, provable it all arrived, and revocable after the fact.
The math and GPU trick behind serving thousands of LoRA adapters from a single base model — without ever merging weights.
Most teams are shipping agents that feel like glorified chatbots - here's what's going wrong and how to actually fix it
How I optimized tool selection for on-device agentic AI — from ~3 seconds to ~15ms using weighted embeddings
Building encrypted vector search with homomorphic encryption — from 6.5s to 172ms with 95% recall
Building an event-driven LLM inference architecture
Learn how to effectively manage stacked branches in Git for cleaner code reviews and faster iteration cycles.