MLOps Framework

An open-source, self-hosted MLOps workspace for computer-vision training, evaluation, model management, and inference workflows. YOLO object detection is the initial reference workload.

Current status

This repository contains two runnable applications:

• A Next.js frontend with live inference and persistent run history integration, alongside prototype screens for dataset and model management.
• A FastAPI inference backend that loads local Ultralytics YOLO models, accepts JPEG or PNG uploads, returns annotated images with structured detections, and persists execution history into a database (SQLite default via SQLModel).

Both the Inference workspace (/inference) and the Run history & details screens (/runs, /runs/details) are connected live to the backend API.

Repository layout

mlops/
├── backend/                  # FastAPI YOLO inference & run persistence service
│   ├── app/                  # Routes, database, domain models, and YOLO inference
│   ├── models/               # Local YOLO model weights
│   ├── README.md             # Backend setup and API reference
│   └── pyproject.toml
├── frontend/                 # Next.js product shell & connected UI
│   ├── src/
│   ├── README.md             # Frontend routes and behavior
│   └── package.json
├── docs/                     # Product plan and design documentation
└── AI_USE.md                 # Project AI-use disclosure

Requirements

Backend

• Python 3.12 or newer
• uv
• Sufficient CPU/GPU memory to load the included YOLO weights

Frontend

• Bun

Quick start

Run frontend and backend as separate applications. Start each in its own terminal.

Backend

cd backend
uv sync
uv run uvicorn app.main:app --reload

The backend is available at:

• API: http://127.0.0.1:8000
• Swagger UI: http://127.0.0.1:8000/docs
• Health check: http://127.0.0.1:8000/health

See the backend README for request examples, supported models, validation behavior, and database configuration.

Frontend

cd frontend
bun install
bun dev

Open http://localhost:3000. The frontend proxies /infer and /inference_runs directly to the FastAPI backend. See the frontend README for available routes and behavior.

Development checks

Run backend checks from backend/ :

uv run ruff check app
uv run ruff format --check app
uv run pytest

Run frontend checks from frontend/ :

bun lint
bun run build

Implemented backend API

Method	Path	Purpose
GET	/health	Report that the API process is running
POST	/infer	Run YOLO inference on an uploaded image & log run to DB
GET	/inference_runs	List all persisted inference runs
GET	/inference_runs/{run_id}	Retrieve details for a specific inference run

Documentation

• Backend setup and API reference
• Frontend routes and mock behavior
• MLOps master plan
• Frontend product-shell design
• AI-use disclosure