Scoping Review AI — Large Text AI Analysis Platform

A production-grade system for automating scoping reviews using LLMs with zero hallucination tolerance.

Architecture

Frontend (Next.js :3000) → Backend (FastAPI :8000) → PostgreSQL+pgvector + Redis
                                    ↓
                           Celery Workers (search, screen, extract, validate)
                                    ↓
                           LLMs (Ollama | Gemini | Groq)

Pipeline

Search (PubMed) → Screen (LLM) → Retrieve (RAG) → Extract (PICO) → Validate (Cross-LLM) → Store → Display

Quick Start

1. Configure Environment

cp .env.example .env
# Edit .env with your API keys:
#   GEMINI_API_KEY=...
#   GROQ_API_KEY=...

2. Launch with Docker Compose

docker compose up -d

This starts 8 services:

• API (FastAPI): http://localhost:8000
• Frontend (Next.js): http://localhost:3000
• Flower (Task Monitor): http://localhost:5555
• PostgreSQL + pgvector: localhost:5432
• Redis: localhost:6379
• Ollama: localhost:11434
• Celery Worker + Beat

3. Default Login

Email: [email protected]
Password: changeme123

4. Pull Ollama Embedding Model

docker compose exec ollama ollama pull nomic-embed-text

Features

Zero-Hallucination Extraction

• All extracted values require source quotes from the original text
• Values without textual evidence are automatically set to "Not Reported"
• Post-extraction grounding validation rejects ungrounded claims

Cross-LLM Validation

• Primary model (Gemini) extracts PICO data
• Validator model (Groq/Llama) independently re-extracts
• Field-by-field comparison generates agreement scores
• Discrepancies flagged for human review

PICO Schema

{
  "population": "...",
  "intervention": "...",
  "comparator": "...",
  "outcome": "...",
  "study_design": "...",
  "sample_size": "...",
  "confidence_scores": {"population": 0.95, ...},
  "source_quotes": {"population": "exact quote...", ...}
}

PDF Processing

• Auto-download from PubMed Central (open access)
• Manual PDF upload via UI
• PyMuPDF text extraction → chunking → embedding → pgvector

API Endpoints

Method	Path	Description
POST	/api/v1/auth/register	Register
POST	/api/v1/auth/login	Login (JWT)
POST	/api/v1/reviews	Create review
POST	/api/v1/reviews/{id}/search	PubMed search
POST	/api/v1/reviews/{id}/screen	LLM screening
POST	/api/v1/reviews/{id}/extract	PICO extraction
POST	/api/v1/reviews/{id}/validate	Cross-validation
GET	/api/v1/reviews/{id}/export	Export CSV/JSON

Full API docs: http://localhost:8000/docs

Tech Stack

Layer	Technology
Frontend	Next.js 15, TypeScript, Vanilla CSS
Backend	FastAPI, SQLAlchemy 2.0, Pydantic v2
Database	PostgreSQL 16 + pgvector
Queue	Celery 5.4 + Redis 7
LLMs	Ollama (local), Google Gemini, Groq
PDF	PyMuPDF
Container	Docker Compose

Development

# Backend only
cd backend
pip install -r requirements.txt
uvicorn app.main:app --reload

# Frontend only
cd frontend
npm install
npm run dev

License

Private — for research use only.