An intelligent conversational health assistant utilizing local Large Language Models (LLMs via Ollama) alongside RAG-enhanced medical knowledge for secure, private, and real-time inference.
Click on this link: https://drive.google.com/file/d/1j1aGap5wFvbiMLGL3qj7H5LAletnLXwS/view?usp=sharing
- Real-time streaming chat: WebSocket-powered, rapid token-by-token responses for a seamless user experience.
- RAG pipeline architecture: Domain-specific knowledge retrieval utilizing ChromaDB and LangChain to minimize LLM hallucination and ground responses in established literature.
- Safety guardrail implementation: Real-time emergency detection, crisis response protocols, and automated symptom urgency triage mechanisms that precede LLM inference.
- Personalized context injection: System prompts are dynamically assembled using the user's explicit medical profile parameters (age, sex, existing conditions, current medications).
- Longitudinal symptom journaling: Comprehensive logging of historical symptoms featuring a zero-shot AI extraction pipeline to ingest unstructured free-text descriptions.
- Secure architecture: JWT-based authentication combined with bcrypt password hashing and localized model execution to guarantee patient data privacy.
| Component | Technology | Description |
|---|---|---|
| Backend API Framework | FastAPI | Asynchronous standard for high-throughput WebSocket endpoints |
| Generative AI Engine | Ollama | Local execution of foundation models (qwen2, llama3, mistral) |
| Retrieval Architecture | LangChain, ChromaDB | Vector search operations for dynamic context provision |
| Client Application | Angular 17 | Standalone component architecture |
| Relational Database | SQLite / PostgreSQL | Structured persistence for user data and symptom ledgers |
| Continuous Integration | GitHub Actions | Parallelized pipeline for linting, testing, and dependency auditing |
- Python 3.12+
- Node.js 20+
- Ollama daemon installed and operational
git clone https://github.com/ansh-varshney/CareBuddy.git
cd CareBuddy/backend
# Initialize and activate the virtual environment
python -m venv venv
venv\Scripts\activate # Windows
# source venv/bin/activate # macOS / Linux
# Provision Python dependencies
pip install -r requirements.txt
# Provision environment variable template
cp .env.example .env# Recommended baseline model for development throughput:
ollama pull qwen2:1.5b
# Recommended model for high-fidelity evaluation:
ollama pull llama3Execute the baseline ingestion script to populate the ChromaDB vector embeddings:
cd backend
python -m knowledge_base.ingestBackend API Service:
cd backend
uvicorn app.main:app --reload --port 8000Frontend Application:
cd frontend
npm install
npm run startAccess the development client at http://localhost:4200. The interactive Swagger API documentation is available at http://localhost:8000/docs.
Execute the automated test suite to validate core functionality and safety assertions.
cd backend
# Provision testing utilities
pip install -r requirements-test.txt
# Execute assertions
pytest
pytest --cov=app --cov-report=term-missingThis repository utilizes GitHub Actions to run the full validation pipeline automatically on pull requests and commits to the main branch.
The pipeline configuration is strictly defined in .github/workflows/ci.yml and currently encompasses:
- Pydantic and FastAPI schema validation (Ruff)
- State-machine validation and regression tests (Pytest)
- Component compilation validation (ESLint)
- Supply chain security auditing (
pip-audit,npm audit)
The system intercepts high-risk natural language tokens prior to routing queries to the generative engine. This rules-based classification guarantees deterministic handling of medical emergencies.
| Trigger Condition | System Response Action |
|---|---|
| Critical Physiology (e.g., chest pain, rapid onset numbness) | Automated directive to contact emergency services (911/108) |
| Psychiatric Crisis (e.g., self-harm, severe ideation) | Automated directive to contact crisis hotlines (988/iCall) |
| Severe Trajectory (e.g., sustained high fever) | Internal categorization as urgency level 4 for triage prioritization |
| Standard Inquiry | Medical disclaimer appended indicating automated, non-diagnostic response |
The CareBuddy framework establishes a baseline pipeline for localized, multi-agent healthcare assistance. Future architectural phases include:
- EHR/EMR Interoperability Layer: Adoption of FHIR and HL7 protocols to permit bidirectional synchronization with standard institutional health records.
- Multimodal Diagnostic Ingestion: Enhancing the RAG pipeline to ingest diagnostic imaging and continuous biomarker data streams from connected wearables (e.g., Apple HealthKit integrations).
- Federated Learning Pipelines: Implementing secure, privacy-preserving federated fine-tuning across localized user instances to improve clinical accuracy without centralizing patient-identifiable information (PII).
- Proactive Anomaly Detection: Applying timeseries forecasting models upon longitudinal symptom journal data to predict and alert users regarding chronic condition flare-ups prior to symptomatic manifestation.
- Provider Triaging Protocols: Export utilities to bundle historic journal data, generated diagnostic assessments, and biomedical profiles into structured clinical summaries for direct transmission to licensed physicians.
MIT License 2026.