API & Personalization Infrastructure
API & Model-Integrated Personalization Infrastructure
Galya is a multimodal taste graph exposed through an API. We compute aesthetic preference from behavioral signals and return structured outputs that systems can use for generation, ranking, and decision-making. This can be integrated anywhere — LLM pipelines, agents, ranking systems, or applications. <br> Galya exposes aesthetic intelligence through a unified API layer — enabling models, agents, and applications to understand and act on user preference. We are accessible from any system that can make an API call.
API-first. Designed to be queried at inference time by models, agents, and systems.
The Problem
Most AI systems personalize using keywords, historical clicks, and demographics. Outputs may look correct, but they are not aligned to what the user actually wants. This leads to regeneration loops, decision fatiugue, and poor user satisfaction. The gap between relevance and resonance causes regeneration loops and poor satisfaction.
The Galya Approach
Galya models aesthetic compatibility between user behavioral signals, entity identity (destinations, products, properties, brands), and relationships in a continuously deepening taste graph.
We embed and cluster multimodal inputs — images, text, audio, video, and behavioral signals — into structured archetype compositions. This enables any system — including agents — to:
Classify entities into structured archetype compositions
Infer user preference structure from minimal input
Compute and return probabilistic compatibility and affinity scores
Enable re-ranking, filtering, or inference enrichment that systems can act opon
Preference is computed before decisions are made and responses are generated.
Core API Surface
All endpoints are authenticated via API key.
Create or register a trackable user object.
Index content, signals, or entities into the graph
Retrieve structured preference composition for a user.
Return entity affinity scores and ranked recommendations.
LLM-compatible endpoint with Galya enrichment tools and structured taste context.
Data Model
Content is clustered into foundational content clusters representing visual, experiential, and sensorial identities. Users are not assigned fixed labels; they exist as positions within a probabilistic preference space.
This shared structure allows compatibility scoring between any user and any entity in the graph.
Taste is modeled as position within a living graph — not a static label.
Integration Modes
Model-Integrated Inference Layer (Fastest Path)
Route LLM calls through Galya's /ask endpoint. We enrich prompts with structured taste context, pass them to your underlying model, and return personalized, preference-aware outputs.
API-First Enrichment Layer
Use /users/create, /index, /users/composition, and /recommend to integrate preference intelligence directly into your existing pipeline with full control over how outputs are used.
Agent-Compatible by Default
Any agent that can make API calls can query the Galya graph. They can traverse preference structures, retrieve aligned entities, and incorporate taste into decision-making. We provide machine-readable documentation, skill files (.md) for agent context, and structured endpoints for traversal.
Use Cases
What Teams Gain
Product Teams
- · Reduced regeneration loops
- · Increased alignment on first result
- · Differentiated discovery experience
- · Faster personalization from minimal data
Data Teams
- · Structured taste taxonomy
- · Archetype-level analytics
- · Compatibility scoring models
- · Multimodal clustering layer
Business Teams
- · Higher conversion rates
- · Increased booking stickiness
- · Stronger repeat behavior
- · Competitive differentiation