Loading...
Loading...
Technical insights on AI systems, product management, scalable architectures, and lessons learned from working on products with 51M+ users at Cafe Bazaar.
We ran our production AI agent as five separate contexts. Unifying them into one architecture changed how I think about agent design: fewer boundaries, better memory, and a simpler mental model for users.
Answers from your own documents are only useful if you can check them. Building a RAG pipeline where every answer carries real citations, and why citations change user trust more than accuracy does.
A chat log is not memory. What I learned shipping a persistent, per-user memory system for a production AI agent: what to store, when to recall, and why forgetting is a feature.
From simple text mining to LLMs: how we used AI at Cafe Bazaar over 10 years, cutting user review response time from 2 days to under 15 minutes and speeding up development with Claude Code.
How we built a high-performance update service at Cafe Bazaar handling 10,000+ requests per second at peak with sub-70ms response times, managing 46GB of data for 51M+ users.
Speaking about open source project governance models at Software Freedom Day in Mashhad, hosted by IEEE Ferdowsi University and GNU/Linux User Group.
My journey as a technical educator—from high schools to professional bootcamps. Teaching Software Engineering with Golang and Python at Quera on behalf of Cafe Bazaar.
Follow me on LinkedIn for more articles on AI systems, product management, and scalable architectures.