🧠 Playground API

A containerized AI-powered FastAPI backend with real-time chat, face detection, and OCR — built using PyTorch, Hugging Face Transformers, and OpenCV. Deployed via CI/CD on a zero-cost cloud setup.

---

📌 Highlights

• ✅ FastAPI + WebSocket + REST — real-time API stack
• ✅ Deep Learning: PyTorch, Transformers
• ✅ Docker-first architecture with optimized layers
• ✅ Deployed on free-tier cloud with secure GitHub-based CD
• ✅ Uses only open-source models — no paid SaaS or API gateways
• ✅ Built with cost-awareness in mind: zero dollar infra

---

💸 Zero-Cost Architecture

🟢 This project is designed to be FREE to run.
All components run on free-tier infrastructure with open-source tooling:

• 🐳 Dockerized deployment
• ☁️ GCP e2-micro VM (always-free tier)
• 🧪 GitHub Actions for CI/CD (free-tier)
• 🧠 Inference using local models — no paid APIs or usage limits
• 🔐 Deployment via secure SSH (no orchestration overhead)

---

🧱 Tech Stack

🔙 Backend

• Python 3.12
• FastAPI (high-performance async API)
• Uvicorn + Gunicorn for ASGI serving
• WebSockets (for real-time chat)
• uv (lightweight virtualenv + dependency manager)

🤖 Machine Learning & Inference

• PyTorch — core runtime for all deep learning models
• Hugging Face Transformers — Qwen2-0.5B-Instruct for chat
• EasyOCR — OCR from images (Torch-based)
• RetinaFace — Face detection with facial landmarks
• OpenCV — Image manipulation & annotation
• Accelerate — Lightweight GPU utilization

---

📦 Features

Feature	Description
/ocr	OCR on images via EasyOCR
/face	Face detection & landmark visualization via RetinaFace
/chat (WS)	Real-time AI chatbot (Qwen2) over WebSocket
/	Root health endpoint
visualize	Returns annotated images on demand (bounding boxes, etc.)

---

🔁 CI/CD Pipeline

• CI: Linting (Ruff) + Tests (Pytest)
• CD: Auto-deploy on Git tag push to GCP VM
• Managed via GitHub Actions:
  • .github/workflows/basic_ci_flow.yaml
  • .github/workflows/basic_cd_flow.yaml
• Secrets Managed:
  • GCP_VM_SSH_KEY
  • GCP_VM_USER
  • GCP_VM_IP

---

✅ Testing

• pytest + FastAPI’s TestClient
• Mocked inference for OCR/Face detection
• Coverage:
  • GET /
  • POST /ocr
  • POST /face
  • Optional visualize=true rendering

---