Things I built while figuring it out.
A mix of AI, NLP, computer vision, and full-stack experiments. Most live, all open source. Click any card to dive in.
Audionaut
End-to-end audio processing in the browser.
A Streamlit app for stem separation and audio enhancement. Splits any track into vocals, drums, bass, and others using HDEMUCS, with a one-click Karaoke Mode and real-time noise suppression powered by DeepFilterNet.
CinemASKit
Personalized movie recommender on top of IMDb's Top 250.
Scrapes the IMDb Top 250 with BeautifulSoup and serves a content-based recommender from genre, keywords, and popularity signals. No login, no friction, just better next-watches.
Text Summarizer
Abstractive summarization with a fine-tuned T5.
Paste text, choose a length, get a clean summary. Built around a fine-tuned T5 with proper preprocessing pipelines so the output stays grammatical even on long inputs.
Native Ad Gen
Multilingual ad creatives, generated on demand.
Type a product, pick a platform and language, get a platform-sized ad image. Supports Hindi, Gujarati, Marathi, Arabic, English with proper Unicode font handling.
Wrapped
Spotify Wrapped, but anytime you want it.
OAuth into Spotify and pull your top tracks, artists, and genres for short, medium, or long term. Beautifully visualized and built to be shared.
Fruit Detector
Real-time fruit detection with YOLOv5.
Custom-trained YOLOv5 for live fruit recognition from images or webcam. Tuned for low-latency inference even without a GPU, useful for retail, agri, and dietary tracking.
BudgetBuddy
Personal finance with OCR receipts and a chatbot.
Full personal finance suite: OCR receipt scanning, EMI and savings calculators, a finance chatbot, and a secure Mongo-backed user system. Mobile-first and modular.
Bulletin Emotion Tracker
Real-time sentiment analysis on the day's headlines.
Pulls live news from NewsAPI across business, tech, sports, and more, classifies each article with TextBlob, and exports clean Matplotlib charts plus an Excel log for trend tracking.
QR-Code-ART-Gen
GAN-painted backgrounds on scannable QR codes.
Combines GAN-generated artistic backgrounds with standard QR matrices via Pillow + OpenCV. Produces visually striking codes that still scan reliably, great for branding and events.
TherapEase
Anonymous, accessible mental health screening.
Quick preliminary screenings for depression, anxiety, and stress. No data is stored. The goal is to lower the barrier to that first moment of self-reflection.
Song RS
Music recs from your top 10 tracks.
Pulls audio features like danceability, energy, and valence via the Spotify Web API and ranks similar tracks with nearest-neighbor search. Feels less like random matching, more like taste.
CV Applications
Real-time face detection + hand tracking.
A small suite of computer vision starters: Haar Cascade face detection and MediaPipe hand tracking with 21 landmarks, designed to be lifted into bigger projects like sign language or touchless UIs.