Constelix

Inspiration

When creating new AI-powered projects, we often lose context by relying on a single conversation with the AI, wasting time and losing context over even small changes in large projects. To improve both the development experience and the visual experience, Constelix introduces a new way of schematically building large-scale projects, making it easier to understand changes and the functionality of the software we create.

What it does

Like an IDE or a dashboard, Constelix first generates a mind map of the entire project. Each part of the project can have its own AI chat, allowing developers to maintain the context of each feature. Every project module can have its own AI agent running in a CLI, and these AI chats can connect with one another as well as with the project’s structural components. Files are visually connected through lines, creating an interactive map of the project.

On one side, it connects AI terminals with each other, and on the other, it displays the relationships between every component. In practice, it can be thought of as a VS Code experience presented through an interactive dashboard.

How we built it

Codex was used to build Constelix from the ground up.

The most important part of the development is the concept of how every component will be interpreted. By carefully defining the project structure, we use AI to generate the project while allowing it to continuously improve itself.

Codex was used throughout the development process to plan the architecture, generate and refactor components, identify implementation issues, create tests, and document the system. GPT-5.6 is used to interpret repository structures, explain relationships between modules, answer project-wide questions, and help developers understand the impact of proposed changes. Through its integrated CLI agents, the AI can also execute self-correction workflows using its own commands, identify implementation issues, and compare the current project with reference repositories to detect architectural similarities, reusable patterns, and opportunities for improvement.

Challenges we ran into

One of our main challenges was designing a system that could represent a large software project as both a visual architecture map and a network of connected AI agents.

Accomplishments that we’re proud of

We are proud that software can evaluate itself through artificial intelligence, improving the overall user experience.

What we learned

We learned how to apply AI in a way that is easier to understand and that can attract more people to use these tools more easily.

What’s next for CONSTELIX

Our goal is to continue evolving Constelix until it becomes a globally accessible platform that helps developers build and understand complex software more efficiently.

Built With

• codex
• dashboard
• python
• vscode