foresight

• 

  Landing Page

• 

  Landing Page

• 

  Dashboard

• 

  More Detail

Inspiration

Engineering teams often struggle to understand the real health of their projects. Issues pile up, pull requests stall, and risks are discovered too late. We wanted to build something that behaves like an engineering manager—someone who continuously observes the project and surfaces problems before they become blockers.

What it does

Foresight is an AI Engineering Manager for GitLab repositories. It analyzes issues, merge requests, commits, and CI pipelines to generate a real-time understanding of project health.

It outputs:

• Project health score
• Key delivery risks
• Bottlenecks in workflow
• Actionable engineering tasks

It can also automatically create GitLab issues based on detected risks.

How we built it

We built Foresight using a lightweight web application connected to GitLab APIs and Gemini models.

The system:

• Fetches repository data from GitLab
• Structures project activity into analyzable signals
• Uses Gemini to reason about risks and bottlenecks
• Generates structured recommendations and actions
• Stores analysis results in SQLite for fast retrieval

The frontend is built as a simple dashboard to visualize project health and recommended actions.

Challenges we ran into

One of the main challenges was converting raw GitLab activity into meaningful engineering signals. Git history is noisy—commits, issues, and pipelines do not directly represent project health.

We had to design a consistent scoring system that balances:

• Development velocity
• Code review delays
• CI failure rates
• Issue backlog pressure

Another challenge was ensuring the AI output was structured and actionable rather than purely descriptive.

Accomplishments that we're proud of

We built a working AI Engineering Manager that turns raw GitLab activity into actionable engineering insights. The system successfully analyzes repositories and generates project health scores, identifies bottlenecks, and automatically proposes engineering tasks that can be executed directly in GitLab.

We are especially proud that the agent moves beyond reporting and actually triggers real workflow actions, such as creating issues based on detected risks.

What we learned

The hardest part was not connecting to GitLab or calling the model, but defining what “project health” actually means in a way that is both useful and computable.

We learned how difficult it is to turn noisy development signals into structured, actionable decisions, and how important it is for AI outputs to be concise and operational rather than descriptive.

What's next for foresight

We plan to extend Foresight into a continuous engineering system that not only analyzes repositories but also integrates team communication tools (like Slack) and project planning systems to close the loop between insight and execution.

Future versions will focus on better prediction of delivery risk and automated sprint planning suggestions.

Built With

• drizzle-orm
• gitlab-rest-api
• google-cloud-run
• google-gemini-api
• next.js
• sqlite
• tailwindcss
• typescript