Moosa Memon
← Index of work

Satellite change alerts, turned into evidence a newsroom can use

Sentinel Dossier: a five-agent pipeline that fuses near-real-time satellite change detection with legal land-use boundaries and multilingual news, reasons about whether a change is legally suspicious, and writes cited, human-reviewable dossiers for the cases that matter most.

Type
Environmental
Stack
Python / Claude API / Retrieval / SVG maps / FastAPI
Status
All planned phases built and tested offline; 51 tests; live data needs only a free GFW API key

Problem

Illegal land clearing, unpermitted mining and concession violations happen continuously and at scale. The raw signals to catch them are public: satellite change alerts fire within days, legal boundaries are mapped, local news surfaces incidents in dozens of languages. No small NGO or newsroom has the staff to correlate all three for every alert, every day.

System

sentinel run ingests alerts, filters and deduplicates them into clusters, spatially joins each cluster against protected-area and concession boundaries, evaluates a legal rule library, retrieves and re-ranks corroborating news, scores violation likelihood, severity and priority, and writes ranked Markdown dossiers with SVG maps. A review dashboard puts a human in the loop. Run it twice and dossiers start noting prior incidents at the same location; run it on a second area of interest and the same pipeline transfers.

Claude is used in tiers: a strong model for high-stakes clusters, a cheap one for routine ones, and a deterministic heuristic when no key is set.

Worth knowing

Have a workflow that looks like this?

Most of these start as a messy, manual process someone got tired of. Twenty minutes is usually enough to sketch how I'd approach yours.