Inspiration
Our team was inspired by the leveraging of AI Tokens to completely rewrite the Javascript/Typescript toolkit, Bun, from Zig to Rust. This rewrite of Bun would've likely cost over a $1 million, yet was able to be finished for only $165,000 in AI API Tokens. This cost-savings demonstrated how AI has the potential to be far more than a code assistant, it can instead be transformed into autonomous engineering workflow. This got our team thinking about how we can use AI Orchestration via Loop Engineering to get different AI Agents to collaborate with one another, essentially team working as two separate Engineers to solve inputted technical prompt cases from users.
What it does
Our project links to users personal AI accounts, such as their Claude and Codex subscriptions. Once the accounts are linked, the orchestrator accepts users prompts. With the prompt submitted, loop engineering autonomously mimics real world software engineering workflows by writing code, writing and running unit tests, and auditing the code. Once the solution passes all the checks it is automatically pushed to GitHub. Along the way a visual interface alerts the user to the progress of answering their request. Providing visual markers of milestones along the way to the final output. By having multiple AI Agents collaborate together, our orchestrator is able to deliver a prompt answer that optimizes token usage.
Challenges we ran into
- Unit Tests written by the Agents would occasionally fail, leading to more debugging to avoid failures.
- Creating an engaging User Interface
Accomplishments that we're proud of
- Building a Full-Stack AI Orchestrator
- Integrating Loop Engineering
Built With
- claude
- codex
- css
- cursor
- html
- javascript
- typescript
Log in or sign up for Devpost to join the conversation.