Hackathon Helper

Inspiration

The inspiration for Hackathon Helper came from a practical realization: every hackathon begins with the same repetitive, manual setup. Whether it's parsing a dense Devpost overview for requirements or initializing a repository with the right .gitignore and requirements.txt, these "cold-start" tasks eat into valuable building time. I wanted to build a tool that treats a hackathon like a professional engineering problem.

What it does

Hackathon Helper is a local multi-agent AI tool that acts as a "First-Hour Co-pilot." By analyzing a hackathon's overview text, it instantly delivers:

1. A Technical Blueprint: It extracts mandatory "must-build" requirements and proprietary platform constraints (like specific API integrations).
2. Dynamic Scaffolding: It generates project-specific configuration files, including .gitignore and requirements.txt, tailored to the specific Python libraries the AI predicts the project will need.
3. Strategic Concepts: It brainstorms 2-3 project ideas that are creative yet achievable, ensuring the developer starts with a valid path to a finished submission.
4. Professional Reporting: Through a local Flask web interface, it renders all findings into a clean, professional dashboard.

How we built it

The project utilizes a LangGraph multi-agent workflow to separate specialized tasks into distinct, logical nodes:

• Analyst Node: Focuses on requirement extraction and constraint mapping.
• Scaffolder Node: Reasons about the project structure and generates local configuration files.
• Creative Node: Brainstorms ideas based on the hackathon criteria.
• Intelligence: Powered by Gradient AI models for high-fidelity text extraction and reasoning.
• Web Interface: A Flask (app.py) server that utilizes markdown2 and custom CSS to transform raw agent data into a polished browser-based report.
• Orchestration: A custom PowerShell layer manages the data flow, feeding raw Devpost data into the environment and capturing the final reports.

Challenges we ran into

The most significant challenge involved managing state across multiple AI agents. Ensuring that the "Scaffolder" understood exactly what the "Analyst" found required careful data modeling within LangGraph. We also faced technical hurdles with character encoding and file-system permissions when automating the generation of sandbox environments. Navigating the nuances of UTF-8 encoding in PowerShell was a secondary but vital lesson in building robust developer tools. Finally, bridging the asynchronous LangGraph logic with a synchronous Flask web server required careful implementation to ensure a smooth user experience.

Accomplishments that we're proud of

We are particularly proud of the Scaffolder Node’s ability to "reason" about dependencies. It doesn't just provide a generic template; it looks at the hackathon goals and suggests the specific libraries needed. We are also proud of the UI/UX layer—moving beyond raw JSON outputs to a professional, browser-based report makes the insights immediately actionable for the whole team.

What we learned

Building Hackathon Helper taught us the power of Multi-Agent Systems (MAS). We learned that breaking a complex problem into specialized agents produces much higher accuracy than a single generic prompt. We also gained deep experience in using LangGraph to manage workflows and learned the value of providing a clean "presentation layer" even for internal developer tools.

What's next for Hackathon Helper

• DigitalOcean Agent Deployment: Finalizing the transition to a cloud-native serverless agent for public access.
• GitLab Integration: Connecting the tool to GitLab webhooks to automatically create project repositories and push the scaffolded files.
• Multi-Language Support: Expanding beyond Python to include scaffolding for Node.js, Rust, and Go.

Built With

• digitalocean
• gradient
• langgraph
• pathvalidate
• python