Blog
System design, written down.
Architecture write-ups on retrieval, agent memory, LLM infrastructure and streaming. Each one starts from requirements, draws the system and lists what breaks.
Detecting leaked code and brand abuse at scale
How to find leaked source code, credentials and brand impersonation on public code platforms without running GPU inference on everything.
Control plane vs data plane: the first split in a video platform
Why the server that decides who can watch must never be the server that delivers the video.
“One million viewers” is four workloads
Control, playback, events and processing grow differently. Scaling starts by asking which one is growing.
The math behind a deterministic agent memory
How an external memory for LLM agents finds connections, orders history in time and picks what to show the model, using only arithmetic.
AI is killing jobs: and that's a good thing
Seven months of job hunting, one brutally honest chatbot, and a list of things I never learned.