I build AI that actually ships.
Computer Science at St Andrews. I turn machine-learning research into full-stack products — and I've shipped them across three internships.
I'm Vansh — a builder who got hooked on code at nine and never really stopped.
These days I'm reading Computer Science at the University of St Andrews (First-Class Honours track, Dean's List two years running), spending my time where machine-learning research meets shipping real software.
Three internships in, I've built everything from LLM-powered platforms to medical-imaging classifiers. When I'm not shipping, I'm at a hackathon, a society meet, or on a cricket pitch.
Three internships, one common thread: shipping.
Built BENPOS, a shareholder-data portal (React, Node.js, SQL) across 4 companies and 300+ client accounts.
Cut record lookup from ~6 minutes to under 20 seconds, and automated ~2 hours of weekly reporting.
Built Python pipelines (Pandas, NumPy) to clean a 50,000+ record dataset of flight and telemetry logs.
Shipped ML tooling (scikit-learn, XGBoost): 91% failure prediction, 88% delay forecasting, K-Means anomaly detection.
Built a booking web app (React, Next.js, Node.js) handling 150+ bookings.
Designed a centralized dashboard that cut manual phone-booking by ~30%.
Things I've built.
An AI-powered, multi-tenant platform for offshore funds with separate investor and manager portals — an LLM-powered due-diligence engine that generates sourced research reports, plus a tenant-isolated AI chatbot with role-based access over sensitive financial data.
Fine-tuned ResNet-18 to detect pneumonia in chest X-rays, using Grad-CAM to interpret attention and catch shortcut-learning failures.
A CNN that flags AI-generated images with a calibrated confidence score, benchmarked against unseen generators for robustness.
A native iOS app (Swift + Firebase) for debt repayment — EMI schedules, payment tracking, Face ID auth, and real-time sync across devices.