Skip to content
Cupikernelhome
APIExplainable matching in one call

The core of every match.

Cupikernel is the matching engine that product teams build on. Send any two entities and get back a similarity score, ranked results and the factors behind them, in milliseconds.

  • Explainable results
  • Configurable logic
  • Privacy by design
31 ms

Role “Senior Backend Engineer, Payments” compared with profile entities. Overall fit of the top result prf_2c91: 0.92. Skills overlap 0.96 at weight 0.40, Experience depth 0.88 at weight 0.25, Domain context 0.93 at weight 0.20, Location & schedule 0.84 at weight 0.15. The final decision stays with your team.

Examples of what teams match: Roles with People, Buyers with Products, Learners with Programs, Founders with Investors, Mentors with Mentees, Tenants with Homes, Projects with Vendors, Volunteers with Causes, Researchers with Grants, Organizations with Opportunities.

How it works

From raw records to explained matches.

Cupikernel sits between your data and your product. You describe the entities, the engine compares them, and every result explains itself.

  1. 01

    Describe

    Model the entities in your domain, such as roles and people, products and buyers, or programs and learners, and stream records in through the API.

  2. 02

    Compare

    Kernel functions measure similarity factor by factor. Your weights combine them, and your rules apply the hard constraints.

  3. 03

    Explain

    Get ranked results in milliseconds, each with a score and a factor breakdown. Your people make the final call.

Core capabilities

One engine for every kind of match.

The building blocks product teams need to ship precise matching without starting from scratch.

  • Similarity kernels

    Compose kernel functions for text, numbers, categories, sets, locations and embeddings. Each factor is measured the way it should be.

  • Explainable scores

    Every result carries its factor-level contributions, so your users and your team can see exactly why two entities fit.

  • Configurable weights & rules

    Tune weights, hard constraints and boosts per use case. Configurations are versioned, reviewable and easy to roll back.

  • Real-time indexing

    New and updated records can be matched within seconds, so results keep up with current inventory, openings and availability.

  • API & SDKs

    A clean REST API with typed SDKs for TypeScript, Python and Go, all generated from one OpenAPI spec.

  • Privacy by design

    Field-level controls, data minimization and regional residency are built in from the start, not added on later.

    Mark a field as excluded and it can never influence a score, no matter how the weights change.

Values

Precise about fit. Careful with people.

The principles behind every decision we make about the engine.

  • 01

    Explainable by default

    Every score comes with its reasons, so no match arrives without an explanation.

  • 02

    Transparent logic

    Weights, kernels and rules are configuration you can read, review and version. Nothing is hidden in a black box.

  • 03

    Privacy by design

    We process only what matching needs, keep it where you choose, and never use it to train shared models.

  • 04

    People decide

    Cupikernel measures fit. Decisions about people and opportunities stay with people.

Build on the core.

Tell us what you're matching. We'll show you how Cupikernel scores it, explains it and serves it at production scale.