Learning Claude Code — via an OTT Streaming Platform

Disclaimer: This project is a learning playground for exploring AI-native platform design. It is not production-ready and not intended for real-world deployment.

A hands-on learning project for exploring AI-assisted development with Claude Code. The vehicle is a real-world OTT streaming platform: complex enough to encounter genuine engineering challenges, concrete enough to produce something you can actually run and interact with.

What This Is

This repo implements a working OTT platform (VOD catalog, EPG grid, live TV, recommendations, entitlements, and more) while deliberately exploring how Claude Code handles real software engineering tasks — not toy examples.

The goal is to learn by doing: each numbered branch (001-*, 002-*, ...) represents a self-contained feature implementation, built with Claude Code as the primary development tool. Browse the branches to see how individual features were approached and built.

New features are developed using speckit — a Claude Code skill that structures feature work through specs, plans, and tasks before writing any code.

Screenshots

Home — personalized hero + content rails	Browse — full catalog with genre & type filters
EPG — 7-day TV guide grid with channel filters	Title detail — play, watchlist, ratings, cast
Profile selector — multi-profile per account	Admin dashboard — catalog, channels, users, embeddings, analytics

Roadmap

See ROADMAP.md for completed features, what's planned next, and overall coverage.

What You'll Find Here

• A fully runnable backend (FastAPI + PostgreSQL + Redis) and two React frontends
• AI-powered content recommendations using pgvector embeddings
• Multi-profile auth, parental controls, entitlements, TVOD, and stream sessions
• Progressive complexity: from simple CRUD to ML-backed features
• Design docs, PRDs, and user stories in docs/ that informed each feature

Prerequisites

• Docker and Docker Compose installed and running

Quick Start

git clone https://github.com/stefanros481/ott-platform.git
cd ott-platform/docker
docker compose up --build

First build takes ~5 min (Python deps + Node deps + sentence-transformers model download). Subsequent starts use cached layers.

URLs

Test Users

Entitlement Tiers

• Basic — 60% of titles, 15 channels
• Standard — 85% of titles, 25 channels
• Premium — 100% of titles, all channels

Seed Data

Data is seeded automatically on first backend startup. To re-seed, delete the postgres_data volume and restart.

EPG schedule stale? The EPG is seeded with ±7 days from the first startup date. If your TV guide looks empty, the schedule has likely expired. To refresh just the EPG without losing other data:

docker exec ott-platform-backend-1 uv run python -c "
import asyncio
from sqlalchemy import delete
from app.database import async_session_factory
from app.models.epg import Channel, ScheduleEntry
from app.seed.seed_epg import seed_epg

async def reseed():
    async with async_session_factory() as s:
        await s.execute(delete(ScheduleEntry))
        await s.execute(delete(Channel))
        await s.commit()
    async with async_session_factory() as s:
        counts = await seed_epg(s)
        print(counts)

asyncio.run(reseed())
"

Tech Stack

• Backend: Python / FastAPI, SQLAlchemy 2.0 (async), Alembic
• Database: PostgreSQL 16 + pgvector (HNSW index for similarity search)
• Cache: Redis 7
• AI: sentence-transformers (all-MiniLM-L6-v2, 384-dim embeddings)
• Frontend Client: React 18, TypeScript, Vite, Tailwind CSS, TanStack Query, Shaka Player
• Frontend Admin: React 18, TypeScript, Vite, Tailwind CSS, TanStack Query
• Auth: JWT (HS256)
• Streaming: HLS/fMP4 via FFmpeg (SimLive), nginx CDN for segment delivery, ClearKey DRM
• MCP Server: FastMCP — read-only AI agent access to the content catalog
• Infra: Docker Compose

Architecture

frontend-client (5173)  frontend-admin (5174)
         \                     /
          \                   /
           +--> backend (8000) <-------+
                |          |           |
           PostgreSQL    Redis    CDN/nginx (8081)
           + pgvector              |
                              hls_data volume
                              (FFmpeg/SimLive)

Single FastAPI monolith with router-based module separation:

CDN / SimLive: An nginx container (port 8081) serves fMP4 HLS segments from a shared hls_data Docker volume. FFmpeg processes (managed by SimLiveManager) write one simulated live stream per TSTV-enabled channel into that volume. The backend's CDN_BASE_URL points manifests at this CDN.

AI Recommendations

Embeddings are generated from title metadata (title, synopsis, genres, cast, mood/theme tags) using all-MiniLM-L6-v2. Similarity search uses pgvector cosine distance with HNSW index.

Surfaces:

• "More Like This" — cosine similarity from a source title's embedding
• "For You" — centroid of user's watched/rated titles → similarity search
• Home rails — Continue Watching, For You, New Releases, Trending, genre rails

Regenerate embeddings (e.g. after changing seed data):

# Via admin dashboard: click "Generate Embeddings" button
# Via API (as admin user):
curl -X POST "http://localhost:8000/api/v1/admin/embeddings/generate?regenerate=true" \
  -H "Authorization: Bearer <admin-jwt>"

Project Structure

├── backend/                 # FastAPI app (Python/uv)
│   ├── app/
│   │   ├── models/          # SQLAlchemy ORM models
│   │   ├── schemas/         # Pydantic request/response schemas
│   │   ├── routers/         # FastAPI route handlers
│   │   ├── services/        # Business logic (incl. SimLiveManager)
│   │   └── seed/            # Data seeding scripts
│   └── hls_sources/         # Source video files for SimLive (per channel)
├── frontend-client/         # Viewer-facing React app
│   └── src/
│       ├── api/             # Typed API client
│       ├── pages/           # Route pages
│       ├── components/      # Reusable UI components
│       └── context/         # Auth context
├── frontend-admin/          # Admin dashboard React app
├── mcp-server/              # FastMCP read-only content metadata server
│   └── src/ott_mcp/
│       └── server.py        # Tools: catalog, EPG, analytics, TSTV, packages
├── nginx/                   # CDN config (serves HLS segments on port 8081)
│   └── cdn.conf
├── docker/                  # Docker Compose + init scripts
├── docs/                    # Design docs, PRDs, architecture, user stories
└── specs/                   # Per-feature specs and validation quickstarts

License

Copyright 2026 Stefan Rosander

Licensed under the Apache License, Version 2.0.