Engineering decisions, product questions and what we learn by building.
I traced a slow ingestion pipeline to CPU embedding inference, evaluated the cost of waiting, and moved that stage to GPUs. A practical account of choosing compute by the cost to finish the job.
Read the articleInside adeka.lk: automated listing review, explainable quality signals and smarter ranking—and a practical route to AI adoption in existing businesses.
Read the articleEvent ordering, position history and the tradeoff between smooth movement and responsiveness.
Read the articleUse shared notifications for logout and tab-local state for messages, while keeping the server authoritative.
Read the articleHow a Bash pipeline, Redis queues and streamed worksheet XML addressed the memory limits of large report generation.
Read the articleA product to build. A team to lead.
A question about where AI fits.