Logo Lanfrica

obooks29/mama-agent-hackathon

Domaine:

healthcare

Type de record:

software
Créateur:
obo
Hôte:
MAMA – Maternal Artificial Multi-step Assistant for malaria in pregnancy in Nigeria # MAMA – Maternal Artificial Multi-agent Assistant *Clinical AI for Nigerian community health workers — powered by Elasticsearch* > **Built for the Elasticsearch Agent Builder Hackathon 2026** --- ## The Stakes A misdiagnosis at 12 weeks gestation — prescribing Artemether-Lumefantrine instead of Quinine — can cause fetal harm. MAMA catches this. Every time. --- ## Problem - Nigeria accounts for **27% of the global malaria burden** - Pregnant women are the most vulnerable; fever could be malaria, UTI, or respiratory infection — symptoms overlap, treatments differ, and a wrong call harms two lives - Community health workers operate at a **1:10,000 ratio** with no decision support - Supply chains are unreliable, with disruption rates as high as **50% in some regions** (e.g., Yobe, North East) - Current diagnostic models exist in research papers — not in the hands of the people who need them --- ## Solution – "Triangle of Reasoning" with Adversarial Internal Review MAMA orchestrates four roles within a single agent (Gemini 1.5 Pro) using Elasticsearch Agent Builder: | Role | Function | Tool Used | |------|----------|-----------| | **Analyst** | Assesses regional malaria burden and supply disruption rate | `get_malaria_trends` | | **Clinician** | Matches symptoms to disease using trimester-specific protocols | Built-in knowledge | | **Logistician** | Verifies local drug availability and flags stock alerts | `check_pharmacy_inventory` | | **Dr. Verify** | Adversarially reviews the draft — always raises at least one challenge | Self-critique | The agent follows a **4-phase workflow**: 1. **Analyze** — call both tools to get regional epidemiological and pharmacy context 2. **Diagnose** — apply trimester-specific clinical protocols 3. **Draft** — form a preliminary recommendation 4. **Review** — step into the Dr. Verify persona, find a flaw, challenge it, then respond with a final defended plan The result is a fully transparent, self-auditing clinical agent …