Jev AI Decision Model
How Jev turns context into decisions
01
What is Jev?
Jev is an AI decision model by TypeSafe AI, built for structured judgments. [1]
- INPUT
Application state
Facts and context
- DECISION
Jev judgment
Typed questions
- ACTION
Code execution
Rules and actions
02
Three ways to ask
Example question
New-country login, then a large withdrawal. Anomalous?
Returns a value between 0 and 1.
Read source ↗Illustrative resultAnomaly likelihood: 94%
Anomalous94%
Not anomalous6%
Fixed teaching data · No live model calls · Probabilities are not guarantees.
Common uses
Classification and routing
Detection and validation
Scoring and ranking
Thresholds and human review
03
Inside this trader
- 1
Market snapshot
Price · Order book · Recent trades
- 2
Choice: buy or sell?
Market state → Side + probabilities
- 3
Check limits and place an order
Risk limits → Price + size → Order
04
Why it can be faster and cheaper
AUTOREGRESSIVE
Generate tokens in sequence
PARALLEL EVALUATION
One state, independent questions
Shared stateStructured results
Noul
Choice
Score
Model speed & cost
Baseline: Claude Sonnet 5 = 1×
| Model / Provider | Relative speed | Chart cost (USD) |
|---|---|---|
| JevTypeSafe AI | 193.6× | $0.39 |
| Claude Haiku 4.5Anthropic | 6.2× | $19.49 |
| Claude Opus 5Anthropic | 2.1× | $176.05 |
| Claude Sonnet 5Anthropic | 1.0× | $117.38 |
| GPT-5.6-LunaOpenAI | 6.0× | $3.31 |
TypeSafe AI chart, not this project's measurements or API prices. Animation differences are compressed; use the labeled ratios. [7]