Skip to content

ClinAgent

ClinAgent is a multi-agent clinical decision-support system built on the Google Antigravity SDK and Gemini 3.5 Flash.

Code Video Cerebral Valley

Core trick

Runs retrieval, imaging, and wearable agents concurrently, then passes their outputs to documentation and verification agents that flag missing data and allergy conflicts.

Tags

healthagents

Description

ClinAgent is a multi-agent clinical decision-support system built on the Google Antigravity SDK and Gemini 3.5 Flash. From a patient's EHR, wearable trends, a clinician's voice dictation, and uploaded medical images or documents, a team of specialized AI agents works in parallel to draft and verify clinical documentation — a SOAP note, prior-authorization request, and discharge summary. Gemini 3.5 Flash's low latency and native multi-modality let every agent run concurrently and reason over images and PDFs directly. Layered safety checks surface missing data and allergy conflicts without ever silently blocking the clinician.
Google-I-O-CV-Hackathon-2026/clinagent★ 12 forksPythonMITCreated May 23, 2026 · last push May 23, 2026

README excerpt

# ClinAgent

**ClinAgent** is a multi-agent clinical decision-support system. Given a patient's
EHR, wearable data, a clinician's bedside dictation, and any uploaded medical
images or documents, it runs a team of specialized AI agents — in parallel — to
produce verified clinical documentation (SOAP note, prior authorization, and
discharge summary) for clinician review.

It is built on the **Google Antigravity SDK** with **Gemini** multi-modal models.

> ⚠️ Research/demo software. Not a medical device; outputs require clinician review.

---

## Highlights

- **Parallel multi-agent pipeline** — retrieval, perception, and wearable analysis
  run concurrently, then feed documentation → verification → safety checks.
- **Gemini multi-modal perception** — uploaded medical **images** (X-ray, ECG,
  photo) and **documents** (lab PDF, referral) are sent to Gemini as raw bytes
  for real analysis.
- **Voice dictation** — the browser's speech-to-text transcribes the clinician's
  notes; the transcript is passed to the model for inference.
- **Wearable / Fitbit trends** — longitudinal vitals (resting HR, HRV, SpO₂,
  sleep, steps, etc.) are charted on the dashboard with a **1 week / 1 month /
  3 month** range selector, and analyzed for signs of deterioration.
- **Layered safety** — a synchronous missing-data check and an AI allergy
  cross-reactivity check. Both are **surfaced but non-blocking**; the clinician
  proceeds with full visibility of warnings and criticalities.
- **Auditabilit

Team