ReproSafe
AI Security & Technical Product Management
A local AI security prototype that turns private Python data-processing bugs into reviewed synthetic reproducers before any AI-assisted patching.
By Shehryar Hassan·2026-09-19
ReactTypeScriptViteExpressWasmerGeminiAnthropic
github.com/shehryaur/reprosafe-demo
The Product VisionHelping teams debug private data pipelines without leaking the raw inputs that caused the failure.
The User Problem
Developers often need help reproducing production data bugs, but sending real fixtures or private code context to external tools can expose sensitive values and customer data.
Architecture
- Accepted a narrow Python module and JSON fixture through strict schemas so every run stays bounded and reviewable.
- Executed cases in an isolated Wasmer worker with no inherited model key and no guest networking.
- Generated deterministic synthetic replacements and executed every candidate before allowing export or optional AI patch review.
Product & Growth Impact
- Reduced accidental data release risk by keeping raw private inputs local by default.
- Turned fragile bug reports into portable repro packages with reviewed source, verifier code, and bounded metadata.
- Bound AI request approval to provider, model, payload identity, and destination so review happens before transmission.