trait-learning

Trait extraction automates learning place characteristics from player sessions and game answers. The Skill integrates Nominatim-derived place data, LLM prompts, and embedding generation to curate and replace existing traits within asynchronous queue workflows, enabling continuous enrichment from gameplay input.

The trait-learning pipeline ingests Nominatim place data, player session descriptions, game answers, and existing trait records. These sources flow through LLM curation and embedding generation to produce enriched trait sets that replace prior traits via a deletion-then-insert replacement policy.

Yes. Trait learning supports both edge function and asynchronous queue workflows. The Skill handles pgmq queue integration and edge-function flow patterns, allowing you to process trait updates in real-time edge contexts or batch asynchronous environments depending on your game's event cadence.

Yes. Trait learning requires embedding generation for each extracted trait. Embeddings enable semantic similarity matching and support downstream filtering, clustering, or retrieval operations within the trait enrichment pipeline.

The replacement policy deletes all existing traits for a place before inserting newly learned traits. This ensures trait sets remain current and prevents stale or duplicate trait records from accumulating across learning loop iterations.

The learning loop captures player session data and game answers at completion, routes them through LLM-based trait extraction and curation, generates embeddings, and applies the replacement policy to delete old traits and insert enriched ones, completing the trait-update cycle.