Hacky Buddy

Inspiration

I’ve spent countless weekends pouring hours into hackathons, only to come away empty-handed more often than not. Watching brilliant ideas go unloved because I simply couldn’t scale my effort across enough contests sparked the idea for HackyBuddy. What if there were an automated partner that could enter—and win—multiple hackathons on my behalf, freeing me up to focus on new projects and real-world challenges?

What it does

HackyBuddy is an AI-powered automation system that:

• Scans upcoming global hackathons around the clock
  
• Generates project ideas tailored to each event’s theme and judging criteria
  
• Builds proof‑of‑concept code, documentation, and polished presentation materials
  
• Submits entries simultaneously to dozens of hackathons in parallel
  
• Optimizes its approach over time by analyzing judge feedback and past winners
  

All of this happens hands‑off, so you can stay focused on learning, networking, or even sleeping—while HackyBuddy works.

How we built it

1. Hackathon Data Pipeline

   • Scraped official hackathon sites and APIs to assemble a real‑time feed of themes, deadlines, and judging rubrics.
     
   • Normalized data into a unified format for easy analysis.

2. Idea Generation Engine

   • Fine‑tuned a transformer‑based model on descriptions of past winning projects.
     
   • Implemented prompt‑templating logic to ensure ideas align with each hackathon’s evaluation criteria.

3. Automated Build & Submission

   • Used CI/CD pipelines to spin up builds in bolt.new, run tests, and package documentation.
     
   • Integrated with hackathon submission portals via headless browser automation and APIs.

4. Feedback & Optimization Loop

   • Captured judge scores, comments, and whether we won.
     
   • Fed data back into our ML models to continuously improve idea generation and presentation style.

Challenges we ran into

• Heterogeneous Submission Processes
  Every hackathon portal had its own quirks: file‑size limits, folder structures, odd naming conventions, and captchas. Building a truly universal uploader took dozens of edge‑case fixes.

• Balancing Creativity & Consistency
  Early on, our AI would sometimes generate wildly creative ideas that didn’t quite match the prompt. We had to refine our prompt‑engineering strategy to keep concepts both novel and on‑target.

• Scale & Reliability
  Submitting to 5+ hackathons in parallel stressed our infrastructure. We had to architect a robust queuing system with retry logic and real‑time monitoring to prevent failed submissions.

Accomplishments that we’re proud of

• Automated Feed
  
• End‑to‑End Pipeline that builds, tests, and packages projects with minimal manual intervention
  
• Feedback Loop driving a measurable jump in win‑rates after just 2 weeks of continuous learning
  
• Launch MVP in under 1 months, from concept to first user test

What we learned

• Raw AI power isn’t enough: prompt engineering and domain knowledge are equally critical to success.
  
• Battle‑testing against real hackathon portals is the best way to harden your pipelines—no amount of local simulation can match live events.
  
• Even the slickest automation needs human oversight early on to catch weird edge cases and ensure quality.

What’s next for HackyBuddy

• Mobile App with push notifications for new hackathons and live status updates
  
• Community Dashboard so users can share tips, custom templates, and success stories
  
• Expanded Integrations for more specialized platforms (e.g., student‑only, industry‑sponsored events)
  
• Advanced Analytics to help users understand which themes, technologies, and presentation styles yield the highest ROI
  

Built With

• bolt-api
• built-with-typescript
• chrome-extension-apis-(manifest-v3)
• elevenlabs-api
• gemini-api
• javascript
• openai-gpt-4
• react-18
• supabase
• tailwind-css
• vite