Back to Projects

TraceLens

An incident triage tool that takes a production log window and points you to the code path most likely involved.

Problem

When a big Java service fails, engineers waste time jumping between logs, dashboards, and unfamiliar code with no clear place to start.

Approach

I built it on top of CodeLens with log parsing, chunk retrieval, call-chain resolution, and LLM summaries tied back to real code paths.

Impact

It turned the first step of incident debugging from a repo-wide hunt into a short, ranked investigation path.

Tech Stack

Python LLMs CodeLens Qdrant Java

Overview

TraceLens sits on top of the code index. You give it a production failure window, and it gives you a plausible code-level trail to inspect first.

Core flow

  1. Ingest a log window around the failure.
  2. Retrieve relevant code chunks and nearby symbols from the CodeLens index.
  3. Resolve likely call paths and supporting dependencies.
  4. Generate a concise investigation summary grounded in the traced code path.

Why it matters

The point is not to pretend the system found the root cause for you. The point is to remove the blank-page problem during incidents. Instead of starting with repo-wide search, you start with a ranked path through the code that already matches the runtime evidence.

Result

TraceLens makes triage faster and less chaotic, especially when the on-call engineer is outside the part of the codebase they know best.