Insights
Notes from the field.
Practical write-ups on building reliable AI products and cloud systems.
Streaming LLM output to mobile with SSE, and surviving malformed JSON
Why Server-Sent Events are a good fit for LLM-powered apps, and a practical pattern for recovering structured data when a model returns broken JSON.
- LLM
- Streaming
- Reliability
Designing an MCP server for private documents: why intent-based retrieval matters
A practical look at structuring a Model Context Protocol server that gives LLMs precise context from PDFs, Word files and spreadsheets, without locking you into one vendor.
- MCP
- RAG
- Architecture