HackHunt
A centralized hackathon discovery platform aggregating competitions from MLH, Devpost, Kaggle, and Devfolio with smart filtering and AI-assisted recommendations.
Ending the Ten-Tab Search
Student developers and hackathon enthusiasts often waste hours jumping between Devpost, MLH, Devfolio, and Kaggle just to find upcoming competitions. Each platform structures its dates, prize details, eligibility criteria, and submission modes differently, making it frustrating to compare opportunities across the ecosystem. I wanted to create HackHunt as a unified aggregator that consolidates coding competitions into a single, organized interface. Developers can filter events by participation format (online or in-person), entry costs, prize amounts, and relevant technical domains without keeping dozens of browser tabs open.
Rebuilding from a Prototype Scaffold
HackHunt originally began as a rapid visual concept scaffolded with Lovable. While the initial prototype had an attractive visual shell, it was powered entirely by hardcoded mock data and lacked any genuine backend or data extraction infrastructure. I took the project down to its foundations and engineered a completely decoupled full-stack architecture. I rebuilt the frontend using React, TypeScript, and Tailwind CSS on Vercel, and developed a dedicated Node.js and Express backend on Render connected to Google Firestore to store and serve normalized hackathon data.
Scraping Four Moving Targets
Aggregating data reliably from four independent platforms proved to be the most technically demanding part of the project. Devpost relies on paginated endpoints, Devfolio renders state client-side, Kaggle exposes custom API structures, and MLH formats schedules inside schema.org microdata. I built dedicated scrapers combining Cheerio for fast static parsing and Playwright for JavaScript-rendered pages. The scraping worker runs daily cron tasks, standardizing inconsistent date strings into ISO timestamps, and committing deduplicated records to Firestore. I containerized the backend using Docker on Render so headless browser dependencies run reliably in production.
Targeted Recommendations with Fast AI
To make exploration more useful than a simple chronological list, I implemented an AI-assisted recommendation engine. Instead of relying on slow cloud models, I integrated Groq's high-speed inference running open-weight language models. The system parses developer queries and profile interests—such as preferred programming languages, hackathon experience level, and topic preferences—extracting semantic intent to rank and recommend matching competitions.
What Web Aggregation Taught Me
Building HackHunt taught me that data aggregation is an ongoing maintenance process rather than a one-time build. External platforms constantly tweak their CSS selectors, modify markup trees, and update anti-bot policies without notice. Handling scraping failures, containerizing Playwright environments, and building schema fallbacks taught me how to write resilient data ingestion pipelines. It demonstrated that automated error recovery and clean fallback states matter just as much as the primary scraping logic.