Spaces:
Sleeping
Sleeping
Hackathon Builder
Update README with Agentic vs Chatbot comparison and new feature highlights
49ca4d3 | title: CodeBlue Context | |
| emoji: 🚨 | |
| colorFrom: blue | |
| colorTo: red | |
| sdk: docker | |
| app_port: 7860 | |
| pinned: false | |
| # CodeBlue Context 🚨 | |
| > *"In the ER, we don't have 15 minutes to read a medical history. When a patient is crashing, we have seconds to make a decision that determines if they live or die. Missing a single drug allergy or a subtle lab trend isn't just an oversight—it's a tragedy."* | |
| > — **The Problem We Are Solving** | |
| **CodeBlue Context** is an explainable, 5-agent clinical emergency intelligence system built for the Agent Assemble Hackathon. It is designed to tackle the most critical, high-stakes bottleneck in modern healthcare: the time doctors waste reading lengthy, scattered patient histories during life-or-death emergencies. | |
| Instead of spending 10-15 minutes extracting vital signs, labs, conditions, and contraindications from a messy EHR chart, CodeBlue **compresses the patient's entire medical context into a structured, highly visual 10-second summary**. | |
| --- | |
| ## 💥 Why This Matters (The Real-World Impact) | |
| Imagine a patient arriving in severe respiratory distress. The physician orders a rapid sequence intubation and pushes a standard dose of succinylcholine. But buried on page 4 of the patient's record is a history of malignant hyperthermia. The patient crashes. | |
| This happens every day. Medical errors are the 3rd leading cause of death. Our system is built to ensure it never happens again by employing **Five Specialized AI Agents** that act as the ultimate safety net for emergency physicians. | |
| ## 🧠 Agentic Orchestration vs. Standard Chatbots | |
| A typical medical chatbot is a single model answering a single prompt. CodeBlue Context is **fundamentally different**: | |
| * **Multi-Agent Pipeline**: Instead of one "black box" answer, we use a **sequential team of 5 specialists**. Each agent (Triage, Guardian, Pathologist, Physician, Attending) performs a specific task and "hands off" its findings to the next, just like a real ER team. | |
| * **Inter-Agent Communication**: Our agents don't just work in isolation; they **cross-reference** each other. If the Pathologist finds a high INR, the Guardian agent automatically blocks anticoagulants in the Action Protocol. | |
| * **Deterministic Guardrails**: We combine LLM intelligence with a **deterministic core**. If the AI is unavailable, our hardcoded medical rules engine still extracts and highlights critical vitals and labs, ensuring the system never fails in an emergency. | |
| * **Explainability**: Every decision includes a "Collaboration Thread" showing the reasoning of each agent, making it a "Glass Box" system that doctors can trust. | |
| ## 🌟 Hackathon-Winning Features | |
| - **Advanced Clinical Consultation**: Beyond the initial brief, doctors can enter "Consult Mode" to ask the 5-agent team specific follow-up questions (e.g., *"Why did you exclude pulmonary embolism?"*). | |
| - **Primary Clinical Directive**: A high-impact banner that identifies the single most important stabilization task in milliseconds. | |
| - **"Hospital Clinical" UI**: A professional, high-contrast "Blue-Green" dashboard with pulse animations, designed for readability in high-stress hospital environments. | |
| - **Real-Time Thinking Logs**: Watch the agents "deliberate" in real-time with terminal-style logs showing their cross-checks and hand-offs. | |
| - **BYOK (Bring Your Own Key)**: Securely inject your own Gemini API key via the UI to bypass server-side rate limits during live demonstrations. | |
| - **Upload Anything (PDF, DOC, JSON, Image)**: Gemini Vision structures raw clinical charts into parsable medical contexts instantly. | |
| - **MCP Compliance**: Fully exposes an MCP Tool Manifest (`/mcp/tools`) for seamless integration with modern agent ecosystems. | |
| ## 🚀 Run Locally | |
| ### Requirements | |
| - Python 3.9+ | |
| - Node.js 18+ | |
| ### 1. Backend Setup | |
| ```bash | |
| python -m venv .venv | |
| source .venv/bin/activate | |
| pip install -r requirements.txt | |
| ``` | |
| Set your API key in a `.env` file in the root directory: | |
| ``` | |
| gemini_api_key=YOUR_GEMINI_KEY | |
| ``` | |
| Start the backend: | |
| ```bash | |
| uvicorn backend.main:app --reload | |
| ``` | |
| ### 2. Frontend Setup | |
| ```bash | |
| cd frontend | |
| npm install | |
| npm run dev | |
| ``` | |
| --- | |
| *Disclaimer: For hackathon and demonstration purposes only. Not for actual clinical use.* | |