JstVerify

When production misbehaves, the evidence is scattered. The user's experience is in a session replay tool, the API activity is in network logs, and the backend story is in a tracing system. Engineers reconstruct what happened by copying identifiers between tools and hoping the timestamps line up.

I named this problem the visibility gap, and it became the core of our positioning. It describes the time, velocity, and trust lost while diagnostic data stays disconnected. The engineering goal that followed was a platform that correlates those signals automatically and preserves context through the whole investigation.

JstVerify runs on serverless AWS, with an AppSync GraphQL API over Python Lambda ingestion workers and DynamoDB. Telemetry SDKs capture DOM mutations on the frontend and distributed traces on the backend.

The founder built the ingestion pipelines, and I built the product on top of them, which meant working at the seams where session events, API logs, and asynchronous trace spans meet. Much of the hard work was keeping session playback aligned with trace data in real time without dragging down browser performance. I also worked below the product layer where features needed it, writing and fixing Lambda code for feedback handling and functional testing, and wiring new resolvers into the CDK application.

The deepest technical surface of the platform is JstVision, an investigation workspace that puts session replay, frontend events, and backend traces on a single synchronized timeline. It began as a prototype I designed, built, and demoed on my own initiative, before it was adopted as the flagship feature.

WILMA is the platform's AI-assisted trace analysis feature. When an error occurs, the system gathers the connected traces, code paths, and state context and generates a debugging analysis a developer can actually act on.

The founder built the model backend. I built the developer-facing side, including how trace context gets assembled for the model and the master-detail panels that present findings next to suggested code fixes. I reused the same context-gathering approach to turn raw AWS security findings into plain-language remediation notes.

On a small engineering team, the founder owned telemetry ingestion, trace processing, and the LLM backend, and I owned the product application, including JstVision, the topology interface, the app shell, and theming. Some features I carried across the whole stack myself, like the structured beta feedback system, which runs from a Python Lambda backend through AppSync to a multi-step review UI. I also wrote the public product website and the positioning it is built around.