Animood

A high-quality anime recommendation engine built with Rust, powered by semantic search and LLM-enriched data. Finds anime based on thematic similarity, story elements, and user preferences.

Overview

This system provides intelligent anime recommendations by:

• Semantic Understanding: Uses Jina Embeddings (small) v2 to capture thematic and narrative similarity
• LLM-Enhanced Data: Each anime entry includes GPT-4o-mini generated insights about pacing, themes, and characteristics
• Context-Aware Retrieval: Synopses are augmented with web search results for richer context
• Hybrid Ranking: Combines semantic similarity with normalized popularity metrics (MAL scores, members, favorites)

Quality Filtering: The dataset contains 6.5k curated anime, filtered to exclude hentai and include only titles with:

• MAL score > 5.0
• At least 4,000 user ratings
• Minimum 10,000 members

I've burnt $2 on openai so if you find useful you can use the data generated by gpt-4o-mini, its in anime-llm-analysis.

Data was collected in late december 2025.

Quick Start

Prerequisites

• Rust 1.70+
• embeddings.bin file (pre-generated with the dataset)

Running the Server

cargo build --release

# Start the API server
cargo run --release
# Server starts at: http://localhost:3005

API Usage

Get Recommendations

curl -X POST http://localhost:3005/query \
  -H "Content-Type: application/json" \
  -d '{
    "query": "psychological thriller with complex characters",
    "k": 15
  }'

Response Example:

[
  {
    "title": "Steins;Gate",
    "score": 0.942,
    "image_url": "https://cdn.myanimelist.net/images/anime/5/73199.jpg",
    "llm_description": "A sci-fi thriller about time travel... Complex characters... Slow-burn pacing..."
  }
  ...
]