Notes from building software.
Practical lessons and technical explorations across full-stack development, backend systems, and applied AI.
Faster Uploads with Background Processing
How background workers can improve document upload responsiveness, with Redis, Celery, reliable task handling, and clear processing status.
· ~7 min read
Reducing Friction in Financial Onboarding
How better workflow design, reliable third-party integrations, and clear verification states helped reduce signup time by 60% and manual review work by 40%.
· ~8 min read
Evaluate RAG at the Retrieval Boundary
How to evaluate retrieval quality, identify missing evidence, and build more reliable RAG systems using practical metrics and structured testing.
· ~10 min read
Hybrid Search: Give Exact Matches and Meaning a Vote
Why combining keyword search and semantic retrieval can improve RAG systems, how reciprocal rank fusion works, and what to measure before adding more complexity.
· ~11 min read
A Larger Context Window Is Not an Evidence Strategy
Why longer context windows don't always produce better answers, and how to select, organize, and evaluate evidence for reliable LLM applications.
· ~10 min read
Tool-Calling Agents Need Explicit Execution Boundaries
How to build reliable AI agents with clear tool contracts, secure execution, controlled retries, and practical safeguards for real-world operations.
· ~11 min read