Our primary goal is to save lives through the early detection of Pancreatic Ductal Adenocarcinoma (PDAC). By identifying the "Metabolic Decoupling" signature—where blood glucose rises while body weight falls—Sentinel provides a critical window for intervention before the disease reaches advanced stages.
PDAC Sentinel is a Clinical Decision Support System (CDSS) designed for General Practitioners (GPs). It bridges the gap between raw longitudinal data and clinical action by integrating deterministic medical protocols with state-of-the-art Large Language Models (LLMs).
The platform follows a robust three-tier architecture designed for reliability and safety:
- Clinical Engine (Deterministic): Evaluates longitudinal data (HbA1c and BMI deltas) to calculate a risk score (0-5) based on validated oncology protocols (END-PAC & UK-EDI).
- AI Briefing Agent (LLM): Utilizes Qwen2.5-Coder-32B via Hugging Face Inference API to synthesize complex metabolic flags into professional executive summaries.
- Safety Gate: A hybrid validation layer that prevents AI hallucinations by forcing deterministic outputs in low-risk cases.
The system monitors specific biomarkers over a 12-month horizon to detect PDAC's paraneoplastic effects:
- Hyperglycemia: HbA1c increase ≥ 1.0%.
- Involuntary Weight Loss: BMI decrease ≥ 1.0 point.
- New-Onset Diabetes (NOD): Focus on patients diagnosed within <36 months.
- Treatment Resistance: Failure of Metformin to stabilize glucose levels in the presence of weight loss.
- Core: Python 3.10+ (Clean Code, Modular Design).
- AI Integration: Hugging Face Inference API with
smolagentsfor structured reasoning. - Data Integrity: Pydantic models for strict validation of clinical inputs.
- Frontend: Streamlit for real-time data visualization and UX.
- Security: Environment-based secret management (No hardcoded API keys).
- Hybrid AI: Combines the reliability of hard-coded medical rules with the synthesis power of LLMs.
- Production-Ready: Developed with CI/CD compatibility and cloud-native deployment.
- Ethical AI: Implements "Guardrails" to ensure high-stakes medical decisions are always backed by data-driven flags.
├── dashboard_sentinel.py # Streamlit UI & Orchestration
├── agent_logic.py # LLM integration & Prompt Engineering
├── clinical_logic.py # Deterministic Risk Scoring Engine
├── schemas.py # Pydantic data models
└── requirements.txt # Dependency management