Welcome to my end-to-end portfolio project!
What youâre seeing: an end-to-end, production-style movie recommender built to demonstrate real engineering choices. You can either click posters to get a Top-5 recommendation list or talk to the âMovie Conciergeâ chatbot for a more interactive, vibe-driven way of finding films youâll actually watch.
How it works under the hood: poster picks prompts feed into a Session-Based Neural Engine. We use a pre-trained Two-Tower model (PyTorch) to generate vector representations for 27k+ movies. When you select films, we build a dynamic "session profile" on the fly (Vector Averaging) to retrieve candidates, which are then refined by a content-aware re-ranker (SVD + MMR) to ensure diversity.
Chatbot as a second recommender: TheMovie Concierge uses a LLM to handle abstract "vibes" and natural language constraints (e.g., "Ghibli but cozier"). It operates independently to brainstorm ideas, then grounds those hallucinations using real-time TMDB data (streaming providers, scores) to ensure suggestions are watchable and accurate.
Why itâs scalable & affordable: the backend runs on fully serverless AWSâAPI Gateway â Lambda (ID mapping + orchestration) â SageMaker Serverless Endpoint (inference). Everything scales to zero when idle to keep cost down, and the UI handles cold starts gracefully instead of pretending they donât exist.
Click movie posters below to add films.