I'm Aman — a CS student at PTU who builds full-stack systems and trains machine learning models that solve real problems.
Currently in my third year of B.Tech (CS & Software Engineering) at PTU, graduating May 2027. I've used my time to go beyond coursework — interning as a full-stack developer, contributing to open source, and building systems that pushed me on performance, scalability, and design.
On the full-stack side, I work across the entire MERN stack — React and Next.js on the frontend, Node.js/Express APIs on the backend, and MongoDB for data. I've built a real-time chat platform using Socket.io, a WebRTC peer-to-peer video streaming app, and a distributed rate limiter using Token Bucket and Sliding Window algorithms. I care about performance: I cut API response time by 15% at my internship by profiling slow queries and adding targeted MongoDB indexes.
In ML, I've built a NanoGPT-style language model from scratch (transformer architecture, attention mechanism, character-level tokenization), a Credit Card Fraud Detection classifier with class-imbalance handling, and a Fake News Detection pipeline using NLP preprocessing and ensemble models. These projects gave me hands-on experience with PyTorch, model training loops, and debugging underfitting vs. overfitting — not just running notebooks.
I'm currently looking for SDE or ML internship opportunities for Summer/Fall 2026. If you're building something interesting, I'd love to talk — reach me at amankumar.cs27@gmail.com or connect on LinkedIn.
Publications
Research publications and academic work will appear here. Currently pursuing undergraduate studies with focus on NLP and full-stack development.
Academic Work
Full Stack Projects
A full-featured e-commerce application with user authentication, product management, cart, and order tracking — built with the MERN stack.
A full-stack blogging platform with server-side rendering, markdown support, user auth, and a REST API backend powered by Express and MongoDB.
A scalable REST API service containerised with Docker, using SQL for data persistence, JWT authentication, and rate limiting middleware.
ML / AI Projects
An NLP-based classifier that detects fake news articles using TF-IDF features and a fine-tuned deep learning model trained on labelled news datasets.
A machine learning pipeline for detecting fraudulent credit card transactions using ensemble methods, SMOTE for class imbalance, and real-world datasets.
A NanoGPT-style autoregressive transformer language model built from scratch in PyTorch — trained on custom text corpora with attention mechanism optimization.
A production-ready sentiment analysis REST API using a fine-tuned BERT model, served via FastAPI with response caching and batch inference support.