Elastic Track · Hackathon Prototype — Research Use Only

The Digital Pharmacist.
Evidence-Backed.

Submit a patient's medical history and proposed medications. Four Gemini-powered agents query the FDA drug label database live via Elasticsearch, evaluate every drug against that patient's specific conditions, and return a full safety report — with verbatim FDA citations — in under 1 minute.

View on GitHub

The Pipeline

How it works

Submit the Case

A clinician uploads a patient's medical history and proposed medications. No rigid format required — the system handles free-form clinical notes, uploaded PDFs, and CSV lab data.

Query the FDA

The ClinicalPharmacistAgent writes and executes ES|QL queries against the fda_drug_labels Elasticsearch index for each drug — retrieving live contraindication and warning text via Elastic's MCP Toolbox.

Receive the Report

Every drug is classified as Approved, Caution, or Contraindicated — each verdict citing the verbatim FDA label text that applies to that patient's specific conditions.

hub

GOOGLE_ADK · SEQUENTIAL

4-Agent Sequential Pipeline

Each agent has a specific task, uses the output of the previous one, and produces a concrete artifact. The pipeline goes beyond answering a question — it reads a database, reasons over patient data, and writes a clinical document.

medication

ProposePrescriptionAgent

Generates a medication list from clinical history alone, when none is provided

optional

science

ClinicalPharmacistAgent

Writes ES|QL queries · retrieves FDA label text · produces Markdown evaluation

ES|QL

data_object

ParseEvaluationToJson

Converts Markdown evaluation → structured JSON with per-drug status fields

description

JsonToHTML

Renders JSON → print-ready clinical HTML report using Tailwind CSS

Live ES|QL Queries

Elasticsearch is not a supporting tool — it's the intelligence layer. The agent has no hardcoded drug knowledge. Every clinical determination comes from a live ES|QL query against the fda_drug_labels index via Elastic's MCP Toolbox.

FROM fda_drug_labels
| WHERE generic_name
    LIKE "*ibuprofen*"
| KEEP generic_name,
    contraindications,
    warnings_and_cautions
| LIMIT 1

format_quote

Verbatim FDA Evidence

Not a generic alert. Not a chatbot answer. The actual warning text from the FDA label, matched to that specific patient's conditions — surfaced automatically by the agent.

format_quote

"Treatment with Ibuprofen tablets is not recommended in patients with advanced renal disease. NSAIDs can lead to onset of new hypertension or worsening of preexisting hypertension."

FDA Drug Label · Ibuprofen · Patient: CKD Stage 4, BP 165/92 CONTRAINDICATED

warning

Why Existing Systems Fail

EHR alert systems are rule-based and fire so many warnings that clinicians click through over 90% of them without reading. The alerts are untrusted because they don't explain their reasoning with patient-specific context.

The gap is not a knowledge problem. It's an information retrieval problem: the right evidence, for the right patient, at the right moment.

The Digital Pharmacist.Evidence-Backed.