Contextual AI Geocoding for Evidence Research
A five-stage pipeline running nine parallel geocoding backends with seven-component scoring, six structured failure modes, and a Spatial Coherence Score.
Five-Stage Pipeline
From raw evidence text to a resolved, scored geometry.
- Extraction — Identifying place-names amidst noisy evidence streams using contextual NLP.
- Querying — Nine services hit simultaneously: Nominatim, Pleiades, WHG, OHM, Geonames, and more.
- Aggregation — Candidate geometries pooled and deduplicated with a seven-component scoring model.
- Disambiguation — Narrative context prunes the result tree to the most spatially coherent candidate.
- Classification — Six structured failure modes distinguish solvable ambiguity from genuine evidential gaps.
Pipeline in Action
Topodex geocoding pipeline being deployed against live evidence corpora.
