Decision Lab

A guide to Jev in production

Build the case for
AI decision automation.

Evaluate Jev for the decisions your operations depend on. Compare costs, examine workflow fit and define what to measure before you change production.

Review a workflow Jev operator guide
01 / Inference cost$0.042

Per million input tokens.
Output tokens are free.

Published pricing ↗
02 / Response time
0.114sJev
vs
8.566sLLM

TypeSafe’s published workflow comparison. Results depend on the task.

View vendor comparison ↗
03 / Decision controlCalibrated
confidence

Route uncertain decisions for review. Validate thresholds on your data.

Understand confidence routing →

The business case

Evaluate one workflow

Start with an illustrative scenario. Replace its assumptions with your operating data.

01

Current operating baseline

For this workflow only.
seconds
End to end, per decision.
Prompt and context. About 4 characters per token.
Include reasoning tokens. Free on Jev.
Use my actual monthly bill instead
USD
Overrides the token-based estimate. Useful with caching or batch discounts.
02

Workflow requirements

Six deployment considerations
03

Confidence routing

Set by the fit check.
These act automatically. The rest go to your current model. A shadow evaluation measures this number.
Running and migration costs
USD
Monitoring, review queue, retries.
USD
Evaluation, integration, rollout.
Usually at or below today’s prompt. No output-format instructions needed.

From a request to an operational decision

A support request becomes a routing decision. Your application receives the selected team, probabilities and a confidence score.

Example from TypeSafe docs ↗

Request

{
  "state": "My running shoes arrived in the wrong
            size. Can I swap them for a size 10?",
  "model": "jev-latest",
  "questions": {
    "department": {
      "type": "choice",
      "instructions": "Which team should handle this?",
      "criteria": {
        "returns":  "Exchanges, wrong or damaged items",
        "shipping": "Delivery status, delays, lost packages",
        "billing":  "Charges, invoices, payment problems"
      }
    }
  }
}

Response

{
  "model": "jev-1.13.0",
  "answers": {
    "department": {
      "type": "choice",
      "choice": "returns",
      "confidence": 1.0,
      "probabilities": {
        "shipping": 0.0, "returns": 1.0, "billing": 0.0
      }
    }
  },
  "usage": { "input_tokens": 328, "output_tokens": 34 }
}
  • choice is always one of your keys. No malformed output.
  • confidence decides the route: act, confirm or escalate.
  • model names the exact version. Pin it and log it.

From estimate to evidence

Validate the case
on your own workflow.

Run Jev alongside your current model. Measure accuracy, escalation rates, latency and cost before deciding what moves to production.

Discuss your workflow