What it does
Nebius LaunchLens is an AI workflow reliability workbench for developers building model-powered applications. It checks whether a model response satisfies the contract an application expects—not just whether the API returned a successful HTTP status.
The workflow
- Define a task prompt and a JSON Schema for the expected response.
- Send a real inference request through Nebius Token Factory using a supported NVIDIA open-source model.
- Parse the response and validate it deterministically against the schema.
- Inspect observed latency, upstream status, token usage when returned, and exact validation errors.
- If validation fails, ask the model to explain the errors and suggest a prompt correction.
- Rerun and compare the new result with the previous run.
Why it matters
A model can return HTTP 200 and still produce malformed JSON or omit fields required by the application. LaunchLens makes these failures visible and gives developers a repeatable test-and-debug loop.
Built for transparent testing
The schema validation is deterministic; the diagnosis is model-generated advice and is labeled separately. Metrics are reported only when measured by the application or returned by the API. The starter keeps the API key on the server and does not persist prompts/results to a database.
Hackathon integration
The intended live integration is Nebius Token Factory plus an NVIDIA open-source model available in the Token Factory catalog. The exact model identifier and live results will be documented after account access and inference are verified. No integration or performance claim should be considered verified until the working demo and public repository are available.
Current status
Prototype foundation and local setup are in progress. A live model call, public deployment, and demo recording remain to be completed.
Built With
- css
- fastapi
- html
- javascript
- json-schema
- nebius-token-factory
- nvidia-nemotron
- python