AI-Powered Matching for Complex Data

Compose filters, vector search, and LLM reranking into a single pipeline. Swap modules, benchmark configurations, and deploy in minutes.

Request Access Read the Docs

Why Velobit?

Composable Logic

Swap embedding models, rerankers, and filter strategies independently
Mix hard SQL constraints with soft semantic signals in a single pipeline
Configure each stage to match your exact business logic

Benchmarking & Iteration

Test different module configurations and measure match quality
Integrate real user feedback to continuously improve results
Prove what drives conversion with data, not guesswork

Portable Infrastructure

Run on managed cloud for instant setup
Export models and indexes to your VPC
Complete data sovereignty and control

How It Works

A three-stage pipeline that progressively narrows and refines matches

Step 1

Index Your Data

Push documents with flexible JSON data. Choose which fields to embed and define filter constraints — we handle the rest.

Step 2

Configure Retrieval

Pick your embedding model and search strategy. We manage vector indexing (HNSW) and hybrid keyword blending automatically.

Step 3

Fine-Tune Reranking

The secret sauce. Train a small, fast model on your interaction data to reorder top results with human-level nuance.

Your matching pipeline, defined in code

Compose stages, swap modules, iterate on match quality

match.ts


const pipeline = new Velobit.Pipeline({
  index: "vendors",
  stages: [
    { type: "sql_filter", query: "revenue > 1000000 AND region = 'EU'" },
    { type: "vector_search", limit: 100, model: "text-embedding-3-large" },
    { type: "llm_rerank", model: "gpt-4o-mini", fine_tuned: true }
  ]
});


const results = await pipeline.match(buyer_requirements);

Architected for complex verticals

Purpose-built for industries where matching matters

Match buyer requirements with supplier capabilities

Index supplier profiles and certifications as documents, then search with buyer requirements. Filters narrow by region, revenue, and compliance while semantic search and reranking surface the best-fit vendors from messy, unstructured data.

Key Capabilities

Filter on hard constraints like geography, certifications, and financials
Semantically match soft criteria like ESG scores and capability descriptions
Rerank results to prioritize the most relevant suppliers

Deploy your way

Choose the deployment option that fits your needs

Managed Cloud

Instant setup
Auto-scaling
99.9% Uptime SLA
Ideal for rapid iteration
Start building immediately

Self-Hosted / VPC

Export your tuned models and indexes
Run within your own perimeter
Complete data sovereignty
No vendor lock-in
Custom compliance requirements