I tracked down an animal abuser using OSINT

This video walks through a text-based semantic search methodology for narrowing a large geolocation search space when traditional visual identifiers fail. The case study involves identifying the exact location of a short video clip taken somewhere in London, a search area with 80,000+ candidate junctions and no distinctive landmarks. The workflow combines OS Open Roads (junction filtering), Google Street View panorama pulling, and CLIP / SigLIP models scored against natural-language prompts describing visible features (railings, road markings, electrical boxes, tactile paving strips). A two-pass architecture is used: ViT-B/16 as a broad first filter across all candidates, then SigLIP SO400M rescoring the top decile with additional directional crops to test spatial constraints (e.g. railings on both sides of a crossing). The workflow could be improved by (1) adding image-to-image matching alongside text prompts where a reference view of the target scene is available, (2) incorporating late-interaction multi-vector scoring for features that occupy small portions of the frame, and (3) offloading the final rescoring pass to cloud GPU compute (e.g. Lambda Labs, RunPod, Modal) to enable faster prompt iteration and larger candidate sets. The back half of the video covers the follow-through after the geolocation was solved; reaching out to nearby veterinary practices, the RCVS Code of Professional Conduct disclosure exception (Chapter 14), and the institutional response from the RSPCA and the identified vet practice. This video is intended for educational purposes, focusing on methodology and analytical reasoning using publicly available data.