Blog
RSSWriting
Notes from building backend systems in production: real-time pipelines, payments, RBAC and the bugs that taught me something.
2026
- Production RAG Pipelines: Retrieval Is a Data ProblemA practical architecture for retrieval-augmented generation that keeps answers grounded as documents, permissions and indexes change.2 min read · LLM, RAG, Vector Search, Backend
- Building Reliable Agentic AI WorkflowsHow to design tool-using AI agents with bounded plans, explicit state and human approval instead of opaque autonomous loops.2 min read · LLM, Agentic AI, Automation, Backend
- Designing MCP Servers That Age WellLessons for building Model Context Protocol servers with narrow tools, clear schemas and predictable failure behavior.2 min read · MCP, LLM, Developer Tools, APIs
- Idempotent Stripe Webhooks, or How to Never Double-Charge AnyoneStripe will deliver the same event more than once. Here's the pattern I use so a retry is a no-op instead of a second order.3 min read · Stripe, FastAPI, PostgreSQL, Payments
- Evaluate LLM Features Before Tuning the PromptA practical testing strategy for LLM applications that catches retrieval, tool-use and regression failures before users do.2 min read · LLM, Testing, Evaluation, AI Engineering
- Real-Time Leaderboards With Redis Sorted SetsGlobal and per-division rankings in O(log n), pushed to clients over WebSockets. Why a sorted set beats ORDER BY for this job.3 min read · Redis, WebSockets, FastAPI, Backend
- Streaming Live GPS With Kafka, Redis and Socket.ioNotes from building a school bus tracker where parents expect the dot on the map to move now, not in thirty seconds.3 min read · Kafka, Socket.io, Redis, PostGIS, Node.js