ls -lt writing/
Writing
Working notes that occasionally grow into essays. Subscribe by RSS.
2026
- How an LLM thinks: from a bedtime story to Q, K and V Start with the version you could tell a five-year-old, then zoom all the way in: tokens, embeddings, attention, how W_Q, W_K and W_V are learned, and what actually happens when a model answers you.
- Lessons from running a multi-agent assistant in production What broke, what we monitored for, and why a simpler agent graph beat a clever one.
- Event streams over cron: rethinking a lead-scoring pipeline Swapping a nightly batch job for Kafka and Lambda cut lead response time from hours to seconds.
- Spring Boot patterns that actually scale The handful of decisions — connection pooling, idempotency, circuit breakers — that matter more than the framework choice.