Africa Modelling Bridge a modelling movement
M10 · DOCUMENT TRIAGE Live · v0.0

Upload a ministry document.
Get research questions in thirty seconds.

Africa Modelling Bridge's flagship AI module. Pick a sample document — a national malaria report, an EPI bulletin, a Francophone strategic plan — and watch Claude draft three candidate research questions a student could actually take up, with methods, data sources and timeline. The magic moment.

Live demo · v0.0 Anthropic API · Claude Sonnet Rate-limit · 3 runs / hour / IP Public documents only
1 · Pick a sampleLive · v0.0
Or upload your own

v0.0 prototype: uploads fall back to the selected sample's analysis. Real document processing arrives v0.5 with editor review.

2 · GenerateClaude Sonnet · Anthropic API

Three candidate research questions, in about 30 seconds.

Pick a sample on the left. We extract the document, identify the indicators and tables, then draft questions at BSc, MSc and PhD tiers — each with methods, data needed and timeline.

PDF
NMCP Annual Performance Report 2024
National Malaria Control Programme · Ministry of Health, Tanzania
01

Public-domain only

Every document AMB processes is already public. The platform makes it findable, citable and teachable — never republishes anything confidential. Hash-checked against a registry of sources MoHs have asked us not to ingest.

02

Editor review is mandatory

Outputs are AI-generated drafts. From v0.5 onwards, every generated question goes into an editor queue. Ernest, country anchors and working-modeller advisers review before anything publishes as a dossier.

03

The cost is small and visible

A single triage run costs about USD 0.06 in Anthropic API tokens. From v0.5 the public analytics dashboard shows month-to-date AI spend per module — funders see exactly what their tokens bought.

WHAT'S NEXT

The full M10 lifecycle — v0.5 onwards.

The demo is the magic moment. The product is the full editor flow: refinement loop, persistent audit log, cross-link suggestions to existing dossiers, OCR for scanned PDFs, ministry-context grounding via embeddings over the indexed corpus.

See the full module status board →
  • Document upload queue + editor inbox Coming v0.1
  • Refinement chat loop · ‘make this MSc-tier’, ‘focus on the Lake region’ Coming v0.5
  • Persistent audit log · 12-month retention then purge Coming v0.5
  • Cross-link suggestions via Vectorize embeddings Coming v1.0
  • OCR for scanned PDFs (PyMuPDF / Mistral OCR) Coming v1.0
  • Public-for-ministry-contacts mode (Cloudflare Access) Coming v1.5