Projects

An advanced scheduling platform similar to Calendly but enhanced with natural language processing capabilities.

A serverless, edge-distributed, multimodal AI chat platform with RAG and a distributed document-processing pipeline.

A study log and streak tracker for system design prep: the LeetCode heatmap and accountability aid, for the part of interview prep that doesn't have one.

A minimalist Pomodoro timer that tracks where your focus time actually goes, synced across devices.

A fault-tolerant distributed storage system that encodes files into fragments distributed across availability zones, with client-side encryption that the server never sees plaintext of.

Autonomous multi-agent research pipeline powered by LangGraph, DSPy, and Gemini.

High-concurrency ticket reservation system with atomic inventory control. Handles the case where hundreds of users try to buy from a limited pool simultaneously: no overselling, no double-booking, no database lock contention.

An AI-powered automation system that scrapes job listings, tailors resumes and cover letters, and achieves 95%+ keyword coverage for ATS optimization.

A predictive ML pipeline achieving 0.871 ROC-AUC in classifying urban food desert status across 800+ census tracts, with an interactive D3.js dashboard.

Explored ways to detect and mitigate deceptive behavior in LLMs using Representation Engineering (RepE) as part of the Apart Lab Fellowship.

A benchmark to assess risks in LLMs across domains like Healthcare, Finance, and Transportation, comparing naive vs. risk-aware completions.

Co-authored a review paper on chatbot applications, challenges, and future research directions, published in Computer Science Review.

A Flask-based web app that predicts diabetes risk based on key health metrics, improving model accuracy to 81% using a Random Forest Classifier.

A framework analyzing AI policies through a risk assessment lens, focusing on ethical and governance challenges across USA, UK, Europe, India, Japan, and Singapore.

Trained CNN and ResNet-50 models to recognize emotions from facial expressions, with a focus on dataset preprocessing and model tuning.

Applied Isolation Forest and Extended Isolation Forest algorithms to detect fraudulent credit card transactions in highly imbalanced data.

A question-answering system using a fine-tuned BERT model on the SQuAD dataset, with optimized fine-tuning techniques and text preprocessing.

Fine-tuned Falcon-7B on therapist-patient conversations to create an AI chatbot capable of providing empathetic responses, deployed with Gradio.

An AI-powered tool to help individuals with Tourette Syndrome communicate more smoothly, using Whisper transcription and attention-based tic removal.