AI-powered analytics

medagent

An LLM agent that queries a healthcare database, generates charts, and cites medical guidelines — all in one reasoning loop.

27
Clinics
578
Diagnoses
31715
Cases recorded
4074
Guideline chunks
Data sources

The agent works across four datasets — two real, one synthetic, and one public reference corpus. Currently active: Uch

🇳🇬 Real data
● Active
UCH Out-Patient Morbidity 2025
University College Hospital, Ibadan — Nigeria

Real out-patient morbidity records from 27 clinics at UCH Ibadan. Each row reports case counts by (clinic, diagnosis, gender, age band).

27 clinics 3,801 records 5 age bands 2025
🇺🇸 Real data
MIMIC-IV Demo
Beth Israel Deaconess Medical Center — USA

De-identified ICU records for 100 real patients. Includes admissions, diagnoses (ICD-9/10), lab results, prescriptions, ICU stays, and vital signs.

100 patients 31 tables 672,772 rows ICU-level detail
🧪 Generated
Synthetic Healthcare Data
Locally generated, simulated Nigerian hospitals

Fully synthetic records for 30,600 patients across 8 hospitals and 7 years. Used for development and testing without privacy concerns.

30,600 patients 57,637 consultations 7 years (2019–2025) No PII
📖 Public reference
WHO Malaria Guidelines
World Health Organization — PDF corpus

WHO malaria treatment guidelines indexed for retrieval-augmented generation. The agent cites specific pages when answering medical questions.

4,074 chunks vector-indexed Cited in answers
Architecture
User
Question
medagent
LLM reasoning loop
sql python rag_search
Tools
What the agent can do
Question type Tools used Example
Data lookup sql "How many patients are there?"
Chart generation sql python "Plot admissions by hospital"
Guideline lookup rag_search "What is the WHO treatment for malaria?"
Compare data to guidelines sql rag_search "Compare our malaria numbers to WHO"
Agent is live. It can query the database, generate charts, and cite WHO medical guidelines. Responses typically take 10–30 seconds per question.
Free-tier note: the first request after a period of inactivity takes ~30 seconds to wake the service.