Extraction: one cheap call per transcript, skip chatter before any LLM, config-driven model, cache geocodes #298

New Issue

2026-10-27T20:03:49-04:00

logan commented

2026-10-27 20:03:49 -04:00

From the 2026-10-27 review (#296). intelligence.py hard-codes gpt-4o-mini (L968) and embeds every scene. The chatter pre-filter should fully gate empty or very short transcripts and routine traffic before any paid call. Geocoding uses paid Google Places/Geocoding on every location with no cache. Add a result cache, plus a local street/intersection lookup later.

Section issues: #160, #242, #104, #7.

From the 2026-10-27 review (#296). intelligence.py hard-codes gpt-4o-mini (L968) and embeds every scene. The chatter pre-filter should fully gate empty or very short transcripts and routine traffic before any paid call. Geocoding uses paid Google Places/Geocoding on every location with no cache. Add a result cache, plus a local street/intersection lookup later. Section issues: #160, #242, #104, #7.

logan added the

type

feature

area/ai-pipeline area/map

priority

p2

labels 2026-10-27 20:03:49 -04:00

logan referenced this issue from a commit

2026-11-08 21:40:57 -05:00

Merge feat/297-cheaper-ai into feat/299-correlation-simplify

logan commented

2026-11-08 21:51:05 -05:00