notion-movies（Notion movies 搜索）

🎬 Notion Movies Skill

This skill transforms a Notion movie database into an intelligent, searchable, and self-enriching system.

It combines:

• Notion (data storage)
• TMDB (movie metadata)
• Ollama (embeddings)
• Qdrant (semantic search)

🧠 Why this skill exists

This is not just about saving and searching movies.

It turns Notion into a semantic knowledge system for movies, allowing agents to:

• understand meaning (not just keywords)
• enrich data automatically
• avoid duplicates
• recommend content intelligently

🔥 What you can actually do with it

🎬 1. Smart movie ingestion

Instead of manually filling fields:

"add Interstellar"

The agent will:

• create the page
• fetch poster
• fetch plot
• fetch director
• fetch genres
• fetch actors, runtime, year
• format everything properly

🧬 2. Automatic duplicate detection

Prevents:

• Interstellar
• Interstellar (2014)
• Interestellar ❌

Agent behavior:

"This movie already exists. Enriching instead of duplicating."

Detection uses:

• fuzzy matching (typos)
• normalized titles (removes year)
• semantic similarity (vector-based)

🔎 3. Human-like search (semantic + explainable)

Instead of exact filters, you can ask:

"movies about time travel"
"romantic movies in Europe"
"movies like Nolan"
"movies that make you think"

The system understands intent, not just words.

🧠 Explainable results

Each result includes:

• similarity score
• reasoning

Example:

{
  "title": "Interstellar",
  "score": 0.92,
  "explanation": "Similar because: genre similarity, plot similarity"
}

This makes results transparent and debuggable.

🤖 4. Personal recommendation engine

Example:

"recommend something like Fight Club"

→ returns semantically similar movies

🧠 5. Turn Notion into a knowledge system

From:

Static database

To:

Living, searchable, intelligent system

⚡ 6. Agent workflows (OpenClaw)

Example flow:

User → "I want something like Interstellar"

→ searchMovies
→ rank results
→ explain results
→ suggest options
→ user selects one
→ addMovie
→ enrichMovie
→ index vector

⚙️ Environment Variables

Required:

• NOTION_KEY
• TMDB_API_KEY

Optional:

• QDRANT_URL (default: http://localhost:6333)
• OLLAMA_URL (default: http://localhost:11434)

🧩 Actions

➕ addMovie

Creates a movie page (basic structure)

{
  "title": "Interstellar",
  "databaseQuery": "Movies"
}

Behavior:

• checks duplicates before creating
• prevents near-identical entries

🎬 enrichMovie

Enriches a movie with:

• Poster (property + cover)
• Plot (Notion blocks)
• Director
• Genres
• Actors (top 5)
• Year
• Runtime
• Emoji 🎬
• Embedding (vector index)

{
  "title": "Interstellar",
  "databaseQuery": "Movies"
}

🔎 searchMovies

Semantic search with explainability

{
  "query": "movies about space exploration"
}

Returns:

• top matches
• similarity score
• explanation

🧬 detectDuplicate

Detects duplicates using:

• embeddings similarity
• fuzzy matching
• normalized titles

{
  "title": "Interstellar",
  "databaseQuery": "Movies"
}

🔁 syncDatabase

Sync Notion → Qdrant (vector DB)

{
  "databaseQuery": "Movies"
}

🧠 Embedding Strategy

Embeddings are built from enriched semantic content:

Title
Genres
Director
Plot
Themes (inferred)

✨ Example embedding input

Title: Interstellar
Genres: Science Fiction, Drama
Director: Christopher Nolan
Plot: A team travels through a wormhole in space...

Themes: space, time, futuristic

⚡ Optimization

• important fields are repeated (weighting)
• themes inferred automatically from plot
• improves semantic recall and ranking

🧠 Data Enrichment Logic

🎥 Director

• Source: TMDB /credits
• Stored in: Director/es

🏷️ Genres

• Source: TMDB /movie/{id}
• Stored in: Género

🧠 Plot

• Source: TMDB overview
• Stored as Notion blocks:

- Callout - Quote

🖼️ Poster

• Source: TMDB
• Stored as:

- Portada - Notion cover

🎭 Actors

• Source: TMDB /credits
• Top 5 stored in Actores

📅 Year

• Source: release date
• Stored in Año

⏱️ Runtime

• Source: TMDB details
• Stored in Duración

🔐 Edge Case Handling

❌ Movie not found

{
  "error": "Movie not found in TMDB"
}

⚠️ Ambiguous results

{
  "error": "Multiple matches found",
  "options": [
    { "title": "Dune", "year": "2021" },
    { "title": "Dune", "year": "1984" }
  ]
}

🧠 Embedding fallback

If embedding fails:

• returns zero vector
• prevents system crashes
• keeps pipeline running

🧬 Duplicate handling

System prevents:

• typos
• alternate titles
• same movie with year variations

⚠️ Notion Schema Requirements

🧩 Extended Schema (Recommended)

🤖 Agent Guidelines

Agents should:

• ALWAYS use databaseQuery
• ALWAYS check duplicates before adding
• prefer semantic search over filters
• explain recommendations when possible
• enrich instead of duplicating
• only fill missing data
• handle ambiguity before proceeding
• gracefully handle API failures

⚡ Performance Notes

• embeddings generated via Ollama (nomic-embed-text:v1.5)
• vectors stored in Qdrant
• each movie indexed as semantic document
• payload includes metadata for explainability

🚀 Advanced Capabilities

• Semantic search (embedding-based)
• Explainable recommendations
• Hybrid duplicate detection (semantic + fuzzy)
• Automatic enrichment pipeline
• Robust error handling
• Agent-ready design (OpenClaw compatible)