Madrigal.co
Contact us

Inference Models

Bring language models into the way your business works

Define how applications request, validate, and use model responses, with explicit routing and operating limits.

Book a meeting

From a business document to structured fields

Define the input, extract agreed fields, and check the response before use.

Input

Business document

Extract the agreed fields from this purchase order and return them in the defined format.

Inference service

Model execution and validation

Structured response

Application-defined format

{
  "type": "purchase_order",
  "number": "PO-0487",
  "supplier": "Example supplier",
  "total": "12450.00",
  "currency": "USD"
}
Illustrative workflow. Actual formats and fields are defined for your use case.

Two core services

LLM Inference

Run a trained language model on new application inputs.

  • Defined request and response formats
  • Output validation and error handling
  • Integration with your existing application
  • Data access and retention requirements

Model service routing

Direct requests to approved model deployments under explicit routing rules.

  • Routing by task or request type
  • Approved alternative destinations
  • Traffic distribution rules
  • Routing configuration and request logs

Decide how the service behaves

Set the service parameters before expanding usage.

  • Response format

    Agree on required fields, output schemas, and validation rules.

  • Latency and capacity

    Set response-time targets, queue behavior, and capacity limits.

  • Failure handling

    Specify timeouts, retries, and what users see when a request fails.

  • Routing rules

    Define approved destinations and the conditions for using alternatives.

A focused project process

From the first use case to rollout planning.

  1. Define

    Identify the task, inputs, and acceptance criteria.

  2. Integrate

    Connect the model service to your application.

  3. Evaluate

    Check representative examples against the agreed criteria.

  4. Prepare

    Document the setup, responsibilities, and rollout plan.

Frequently asked questions

Which models can you work with?

We assess model and provider compatibility against your requirements.

How do you handle sensitive data?

We define allowed inputs, approved destinations, and access requirements with your team.

What happens if a model request fails?

Agree on retry, fallback, and error-response behavior before rollout.

Define your first inference use case

Share the task, expected response, and application you want to integrate.

Book a meeting