AI Video Livestream Recommender Platform

Budget: $150 - $200

I am building a live-streaming hub where viewers receive real-time, personalised video suggestions while they watch. The heart of the product is an AI engine focused squarely on content recommendation, starting with YouTube streams. As users join, watch, pause or chat, the system should instantly refine what it serves next, keeping engagement high without interrupting playback.

Core workflow • Ingest live and upcoming YouTube streams through the official API. • Analyse viewer behaviour (watch time, chat keywords, reactions) and channel metadata on the fly. • Run a lightweight model—think collaborative filtering plus contextual embeddings—to score and surface the next best video. • Present recommendations unobtrusively inside the player and in a dynamic side rail. • teractions for continuous model retraining and audience analytics.

I will provide API keys and a basic wireframe; you connect the data layer, model, and front-end placement. If you prefer to architect with a micro-service approach (Python/Node for data, React/Vue for the UI), that works, but keep the stack readable and container-ready so I can scale later. Twitch and Facebook Live integration is planned for phase two, so design with those endpoints in mind even if you do not wire them up yet.

Deliverables

1. Source-controlled codebase (Git) with Docker files.
2. Working MVP deployed to my test server showing live YouTube feed plus AI recommendations.
3. README covering setup, model choice, and where to plug new providers.
4. Brief performance report: latency from user action to new recommendation and memory/CPU footprint under 1 k concurrent viewers.

Acceptance criteria • Average recommendation latency under two seconds on a 720p stream. • Precision@5 ≥ 0.6 on a test set drawn from my sample channels. • Clean hand-off call to add future providers without touching core logic.

Once this milestone is stable, I will move on to the deeper analytics layer and cross-platform rollout.