Important

This application is an early preview. It may not run stably and extracted results can be inaccurate. Always check outputs for validity before using them in practice.

LLMAIx (v2) Web

A modern web interface for the LLMAIx framework that turns unstructured medical documents into structured JSON with privacy‑first controls. The application includes an admin dashboard for configuring providers, prompts, schemas, and overrides, plus basic user management (e.g. invitations when enabled).

---

✨ Features

• Upload & organize: multi‑format file support (PDF, DOC/DOCX, PNG/JPEG images, CSV/XLSX) with column selection, grouping and previews.
• Preprocessing & OCR: pluggable methods including Tesseract, Marker, PaddleOCR, and vision LLMs; document parsers like Docling, PyMuPDF4LLM, and MarkItDown.
• Visual schema editor: tree‑based editor supporting nested objects/arrays, all JSON types, import/export, validation, and templates.
• LLM trials: run multiple extraction trials across prompts/models with temperature control, per‑trial iterations, token tracking, and OpenAI‑compatible endpoints (official OpenAI, vLLM, llama.cpp/Ollama, custom gateways).
• Evaluation: upload ground truth, compare field‑by‑field, compute overall/per‑field accuracy metrics.
• Privacy‑first: run fully locally or with self‑hosted providers; no forced external calls.
• Admin dashboard: provider configuration, user management (setup first admin, invitations/roles).

Tech stack: Vue 3 + Vite + TailwindCSS (frontend), FastAPI (backend), SQLAlchemy, Celery, Pydantic for configuration.

---

📦 Requirements

• Node.js 18+ (for the frontend) and PNPM/NPM/Yarn
• Python 3.11+ (for the backend)
• Docker & Docker Compose (recommended for deployment)
• Optional GPU stack: NVIDIA driver + NVIDIA Container Toolkit
• OpenAI-compatible API (e.g., official OpenAI, self-hosted llama.cpp, vLLM, …)
• (Optional) MinIO for object storage or any other S3 storage

---

🐳 Docker Installation

1. Copy and edit environment files:

cp .env.example .env
cp backend/.env.example backend/.env
cp frontend/.env.example frontend/.env  # if present

2. Start the stack (choose GPU or CPU):

# GPU (requires NVIDIA driver + container toolkit)
docker compose -f docker-compose.gpu.yml up -d --build

# CPU
docker compose -f docker-compose.cpu.yml up -d --build

Notes:

• The GPU stack reserves an NVIDIA device via Compose deploy.resources.reservations.devices.
• Services included: backend, worker_default, worker_preprocess, frontend, postgres, redis, minio, and one‑shot minio-init (creates S3_BUCKET_NAME).
• The backend reads settings from backend/.env (see ENV_PATH in backend/src/config.py). The project .env is mounted into /app/backend/.env in the container.

3. Visit the web UI at http://localhost:5173.

---

🚀 Get started

1. Open http://localhost:5173.
2. Create an Admin account.
3. Log in with your new account and configure an LLM provider under Settings.
4. Upload a few documents, define a schema, and run your first trial.

---

🔧 Development

Initialize users

python -m backend.scripts.populate_users

When using Docker

## Replace gpu compose file name with cpu if needed
# Running stack
docker compose -f docker-compose.gpu.yml exec -it backend \
  python -m backend.scripts.populate_users

# Stopped stack
docker compose -f docker-compose.gpu.yml run --rm -it backend \
  python -m backend.scripts.populate_users

MinIO (optional for dev)

Start on macOS:

minio server miniodata

Run backend tests

ENV_PATH=backend/.env uv run pytest --verbose --cov=backend --cov-report=html

---

🔐 Security & Privacy

• Keep PHI strictly local unless you explicitly configure a remote provider.
• Prefer self-hosted, OpenAI-compatible endpoints for clinical data.
• Review your .env secrets and never commit them.

---

📜 License

This project is licensed under the AGPL-3.0. See LICENSE.