Blog
Thoughts on AI, LLM systems, on-device ML, and software engineering.
-
The Voice Agent Problem Nobody Talks About
When you talk to a voice agent, the hard part is not transcription or synthesis. The hard part is what happens when you interrupt it mid-sentence, or pause to think and then keep talking, or when t...
Read more → -
How Claude Learned to Watch Videos: The Decomposition Pattern for Multimodal AI Agents
Claude can read a webpage. It can run a script. It can browse a repository. What it cannot do, out of the box, is watch a video. You paste a YouTube link and it has to either guess from the title o...
Read more → -
How Alibaba's OpenCodeReview Achieves 9x Token Savings: A Deep Dive into the Hybrid Architecture
When Alibaba open-sourced their internal code review tool last month, the numbers caught my attention. 12,000+ stars in weeks. Claims of “higher precision and F1 than general-purpose agents” at “~1...
Read more → -
Building Code Buddy: A Multi-Agent GenAI Code Review Platform at Qualcomm
At Qualcomm, I had the opportunity to architect Code Buddy, a multi-agent GenAI code-review platform that’s now deployed across 4,000+ repositories and serves 12,000+ engineers. Here’s the story of...
Read more →