DevLens AI Autonomous Engineer

Inspiration

Modern software systems generate complex runtime crashes and logic bugs that disrupt developer velocity and increase MTTR (Mean Time to Resolution). Managing production errors manually often requires hours of parsing logs, writing reproduction tests, and creating hotfixes. We built DevLens to transform crash resolution into a fully autonomous, background workflow.

What it does

DevLens is an autonomous AI software engineer designed to detect, verify, and resolve application bugs automatically:

• Real-time Crash Detection: Listens to runtime crash webhooks as errors occur in the application.
• Context-Aware Reasoning: Reads plan-first architecture documentation (scope.md, prd.md, spec.md, checklist.md in the devpost/ directory) to adhere strictly to project guidelines and boundaries.
• Autonomous Remediation: Parses crash logs, analyzes the codebase, and reproduces the error using test suites.
• Automated Resolution: Applies code patches, verifies fixes against test suites, and opens a fully documented Pull Request on GitHub.

How we built it

• Architecture: Built around a modular AI agent pipeline that ingests webhook payloads and project specifications.
• Test Verification Environment: Integrates directly with local testing frameworks (such as Jest) to validate bug reproduction and fix correctness before pushing changes.
• Version Control: Uses GitHub APIs to automate branch creation, commit staging, and PR submission with detailed context logs.
• Plan-First Design: Managed project requirements using structured documentation files (devpost/ directory) to keep the AI agent grounded within defined scope parameters.

Challenges we faced

• Prompt & Context Alignment: Ensuring the AI agent strictly respected project boundaries specified in devpost/ files without hallucinating unnecessary code refactors.
• Test Cycle Reliability: Creating a tight loop where local test environments consistently captured error states and confirmed resolutions before triggering Git commits.

Accomplishments that we're proud of

• Successfully created an end-to-end pipeline where a runtime crash automatically results in a verified, ready-to-merge GitHub Pull Request without human intervention.
• Built a clean, language-agnostic architecture capable of handling runtime errors across diverse applications.

What we learned

• The importance of a "plan-first" methodology when orchestrating autonomous AI agents.
• Best practices for binding LLM reasoning engines with deterministic local test harnesses and continuous integration tools.

What's next for DevLens

• Supporting multi-file complex dependency refactoring.
• Integrating multi-agent peer reviews before opening Pull Requests.
• Adding support for additional CI/CD pipelines and monitoring services like Sentry and Datadog.

Built With

• ai
• autonomous-agent
• devtools
• github-api
• javascript
• jest
• llm
• node.js
• webhooks