MediProc-AI is a high-performance, multi-agent medical intelligence system designed to unify siloed clinical data. It transforms raw medical documents (images/PDFs) into actionable clinical insights using a Triple-Database Hybrid RAG architecture.
Medical data is often trapped in inconsistent formats. Traditional systems struggle to combine semantic meaning (what a patient says), structured facts (lab results), and clinical context (historical relationships).
MediProc-AI solves this using three specialized data stores:
- Weaviate (Vector): Semantic search engine for natural language discovery.
- NebulaGraph (Graph): Knowledge graph for mapping patient-finding-history relationships.
- PostgreSQL (Relational): The clinical source of truth for high-fidelity structured data.
- Language: Java 23 / Spring Boot 3.5.9
- AI Orchestration: Spring AI / LangGraph4j (7-Agent Workflow)
- Databases: Weaviate, NebulaGraph, PostgreSQL
- Ingestion: OCR (Tesseract/OpenCV)
- LLM: OpenAI GPT-4 / local ONNX Embeddings
The system utilizes 7 specialized AI agents to process records:
- Supervisor: Quality control and routing.
- Extraction: Identifying clinical entities.
- Context: Analyzing negations and timeframes.
- Validation: Cross-referencing findings.
- Medical Coding: ICD-10/SNOMED-CT mapping.
- Explainability: Evidence citation.
- Synthesis: Generating clinical executive summaries.
- Full Analysis:
POST /api/v1/process/full-analysis/{patientId} - Semantic Search:
GET /api/v1/query/search/semantic?query={text} - Graph Network:
GET /api/v1/query/patient-graph/{patientId} - Clinical SOAP Recommendation:
GET /api/v1/query/final-recommendation?query={text} - Architecture Demo:
GET /api/v1/query/system-architecture-demo
- Automated SOAP Notes: Generates standardized Subjective, Objective, Assessment, and Plan documentation.
- Cross-Database Discovery: Identifies comorbidities in the Knowledge Graph that semantic search might overlook.
- Triage Prioritization: Automatically flags CRITICAL cases based on extracted vitals and history.