Some of the coolest things I’ve built under time pressure, caffeine, and the thrill of hacking with a team.
April 2025 | Xubuntu, LXQt, Python, Kolibri, ZRAM, Shell Scripting
Built during a 24-hour hackathon, EduLite OS is a custom Linux distro crafted for students and schools with limited hardware access. Based on Xubuntu and optimized with LXQt, ZRAM, and minimal background services, it runs beautifully on machines with ≤2GB RAM. It comes pre-installed with offline learning tools like Kolibri, Python educational apps, and custom teacher dashboards. We even designed a future-ready UI for AI chatbot integration, exam/test mode, and voice support.
May 2025 | Next.js 15, TypeScript, Tailwind CSS, Supabase, Prisma, Groq, Google Search API
Built in a 24-hour hackathon, Aryabhatta Search is an AI-powered platform that simplifies learning discovery for users across all academic levels—from kindergarten to post-graduation. It uses Groq's LLaMA 3 to generate personalized, structured summaries based on user age and educational background, and recommends books and resources using TF-IDF on past search history. The platform integrates Google Custom Search for broader results, supports community forums and parental controls, and was built using a modern full-stack setup with Next.js, Supabase, Prisma, and Tailwind CSS. BTW we win this hackathon at innovate you techathon.
April 2025 | Python, SciBERT, Sentence-BERT, Scikit-learn, Transformers
Developed during a hackathon, this project predicts whether a research paper is publishable and recommends the most relevant top-tier conference, including CVPR, NeurIPS, EMNLP, TMLR, and KDD. It leverages SciBERT for generating contextual embeddings, a self-training classifier for publishability assessment, and Sentence-BERT for similarity-based conference matching. The tool processes labeled and unlabeled paper datasets and outputs predictions in CSV format, offering a practical assistive tool for early-stage researchers. We secured first prize in this hackathon held at Army Institute of Technolgy,Pune.
March 2025 | Python, PuLP, Pandas, Docker, YAML, Data Engineering
This project provides a modular system to automate fantasy cricket team selection using historical and recent player performance data. It calculates form scores via exponentially weighted averages of batting, bowling, and fielding metrics, normalized to a 0–100 scale. Using these scores, the system optimizes an 11-player fantasy team via linear programming (PuLP), applying constraints like role quotas and leadership multipliers (captain, vice-captain). The architecture supports CLI and Docker usage, with a clean codebase powered by a YAML-configured pipeline.