Platform

One engine behind every product

Every Geny product runs on the same AI and application platform. It lets a small team build, ship and operate several production products in regulated domains without rebuilding the foundations each time.

What is inside

The layers every product reuses

Each layer is built once, hardened in production and shared across products.

Model-agnostic LLM layer

One provider abstraction across Anthropic Claude, Azure OpenAI, OpenAI and open models, with tier-based routing, structured JSON output and vision input. Each task gets the right model, with no lock-in.

Retrieval and grounding

Retrieval-augmented generation over curated knowledge bases, so answers are grounded in sources instead of guesses. In genyHealth it runs a dual-embedding setup over clinical and genomic knowledge.

Document AI

Vision OCR and LLM extraction for receipts, invoices and long documents, with chunked processing and validation against business and fiscal rules.

Agents and MCP

Tool-calling agents and custom Model Context Protocol servers that let AI act on real systems, step by step, with every action logged.

Application platform

Multi-tenant organisations, signed-token authentication, role-based access, billing, notifications and audit logging, shared by every product.

Built for real markets

English and Romanian from day one, automated tests and end-to-end browser checks on every product, and deployment practices reused across the portfolio.

How we build

Accuracy and auditability where it matters

Tax, health and business decisions leave no room for confident guesses. The platform is designed around that.

  • Validate, do not trust

    AI output is checked against rules, schemas and source data before anyone relies on it.

  • Humans approve what matters

    Approval steps and reviewable reports keep people in control of consequential outcomes.

  • Every action leaves a trail

    Significant actions are logged, so teams can see who did what, and when.

  • Right model for the job

    Because the model layer is provider-agnostic, we can change models as they improve without rewriting products.

Need this engine for your own product?

We design, build and operate production AI for teams that need it to work in the real world.