LANRE025/Luminalix

Luminalix Outbreak Vulnerability Sentinel — An AI agent built on DataHub that closes the gap between slow, expensive disease surveillance surveys and real-time reality. Cross-references stale survey data with hospital admissions trends and resource allocation to flag high-risk, under-resourced regions. # Luminalix An agent that detects **blind spots in global disease surveillance** by cross-referencing three signals per region: 1. **Official survey data** — slow, authoritative, goes stale 2. **Hospital admissions data** — fast-moving proxy signal 3. **Resource allocation data** — funding / staff / vaccine stock The agent flags regions where survey data is stale **AND** admissions trends suggest an emerging risk **AND** resources are inadequate, writes a justified vulnerability assessment back into DataHub as an annotation, and serves the results through a React web app. Built for the **Build with DataHub: The Agent Hackathon**. --- ## Testing without local setup No GCP project or DataHub instance needed to see the system in action: - **Demo video**: *(placeholder — link to your demo recording here)* - **Sample outputs in `examples/`**: - `examples/vulnerable_regions_report.json` — the real output of a full 30-region agent run (30 evaluated, 18 flagged) - `data/latest_report.json` — most recent live run of the pipeline - **DataHub UI screenshot**: `examples/datahub_writeback_findings.png_1` and `examples/datahub_writeback_findings.png_2` — show real Luminalix findings (including High-vulnerability Bauchi, Ondo, and Ituri) written to the `regional_survey_data` dataset's Documentation tab To run it yourself, follow the setup steps below — you will need a Google Cloud project with the Vertex AI API enabled. --- ## Architecture ``` DataHub Instance ◄────► MCP Server (mcp-server-datahub) │ ▼ FastAPI Backend ──► data/latest_report.json (agent orchestrator, (JSON persistence) Gemini reasoning, REST endpoints) │ JSON / REST ▼ TypeScript React frontend (regions list + detail view) ``` - **Backend**: Python + FastAPI (`backend/`) - **Reasoning**: Gemini via Google Vertex AI (`google-genai` SDK, `backend/app/agent/reasoning.py`) - **Data platform**: DataHub (GMS v1.7.0) via the DataHub MCP server — dataset-level schema / freshness / lineage reads **and* …