Developer Builds AI Agent to Automate Tracking San Francisco Civic Meetings
A developer created Argus, an AI agent designed to monitor San Francisco's civic processes across multiple publishers and formats, automating tasks like agenda tracking, calendar updates, and email notifications. The tool follows issues through seven steps of city government activity — from spotting agenda changes to logging hearing outcomes — that would otherwise require near part-time effort to manage manually. During development, the builder encountered a key retrieval failure where Argus could not surface a known fact about the city's mayor despite having the answer stored in its database, exposing a fundamental flaw in keyword-only search. The fix involved running both keyword and vector searches simultaneously and merging results using reciprocal rank fusion, which avoids the problem of a weak keyword match blocking the more semantically aware vector search. A calibrated distance threshold was also set using real query data to prevent the agent from returning confident but irrelevant results to off-topic questions.
