Writing
Technical writing on decisions made, tradeoffs accepted, and things that shipped.
- How I integrated an LLM into SahamLens without leaking portfolio data.#sahamlens#llm#anthropicRead post
How I integrated an LLM into SahamLens without leaking portfolio data.
A two-stage pipeline: local summarisation first, LLM analysis second. The portfolio data never leaves the machine. Here is how the architecture works and what the prompt engineering problem actually was.
May 25, 20263 min read - I picked DuckDB over SQLite for SahamLens. Here is what the trade-off cost me.#sahamlens#duckdb#pythonRead post
I picked DuckDB over SQLite for SahamLens. Here is what the trade-off cost me.
SahamLens runs on DuckDB, not SQLite. The 14× speedup on analytical queries was the easy part. The concurrent connection bug in the dev server was not.
May 20, 20263 min read