Phase 1 – TMDB Chatbox¶

Goals¶

• Natural-language → structured TMDB queries
• Dynamic endpoint selection (/discover/movie, /person, etc.)
• Multi-entity joins (e.g., “movies starring A and B”)
• Response formatting: list, timeline, comparison, fact

Architecture¶

flowchart TD
  Q[User Query] --> C[Intent/Entity Extractor]
  C --> R[Hybrid Retrieval (semantic+symbolic)]
  R --> P[Planner & Constraint Engine]
  P --> E[Endpoint Executor]
  E --> V[Post-Validation & Rerank]
  V --> F[Formatter (templates)]
  F --> A[Answer]

Repo¶

• https://github.com/FinnMacCumail/tmdbGPT

Highlights¶

• ChromaDB semantic routing: 54 TMDB API endpoint descriptions for intelligent query-to-endpoint matching
• Entity resolution via TMDB Search API calls (people, genres, networks/companies) with parameter mapping
• Constraint-based validation & reranking; flexible response templates