Back to Home

My Projects

A collection of AI, machine learning, and software engineering projects showcasing innovation and technical expertise.

Machine Learning

ETL Pipeline
RAG • Vector Search • AI

Clinical Trials RAG Pipeline

Production-ready ETL pipeline with RAG capabilities for clinical trials data. Built with PySpark + Delta Lake for data processing, Great Expectations for validation, OpenAI embeddings with FAISS vector search, and multilingual GPT-4 query interface. Features comprehensive audit trails and enterprise-grade architecture.

PySparkDelta LakeOpenAIFAISSGreat ExpectationsRAGETLPythonGPT-4Vector Search
Cancer Detection
ML • Healthcare • FDA

Early Cancer Detection ML System

Developed FDA-compliant ML models for early cancer detection at Memorial Sloan Kettering. Achieved 92% sensitivity and 88% specificity on EHR data from 15K+ patients. Implemented SHAP explainability framework for clinical interpretability and conducted external validation across 3 hospital systems.

Pythonscikit-learnSHAPFDA ComplianceEHRClinical MLExternal ValidationAUC/Brier Scores
Robotic ML
EMG • Real-time • Robotics

ML for Wearable-Robotic Communication

Developed real-time ML system for EMG-based wearable-robotic communication at AIR LAB. Improved system reliability by 25% and reduced latency by 40%. Implemented custom signal processing pipeline handling 1000+ Hz biosensor data with sub-10ms response time.

EMGReal-time SystemsRoboticsSignal ProcessingWearablesBiosensorsLow Latency

Research

Material ML
Physics • Research • POSTECH

Physics-Enhanced ML for Material Prediction

Developed physics-informed ML models for yield strength prediction in metallic materials at POSTECH. Achieved 15% improvement in prediction accuracy over traditional methods. Published in Acta Materialia (Impact Factor: 9.4) with 50+ citations.

Physics-informed MLMaterial SciencePOSTECHActa MaterialiaYield StrengthMetallic Materials
BodyTrak
Computer Vision • Privacy

BodyTrak: Privacy-Preserving Pose Estimation

Developed full-body pose estimation system using miniature wrist camera for privacy-preserving body tracking. Published in ACM IMWUT (Impact Factor: 7.9). Custom CNN architecture achieved real-time performance with 95% accuracy on pose keypoints while maintaining user privacy.

Computer VisionCNNPose EstimationPrivacyReal-timeACM IMWUTWearables

Natural Language Processing

AI Coaching
GPT-4 • RLHF • Career

ACES: LLM-Powered Job Coaching Platform

Developed personalized job coaching system using GPT-4 and RLHF at Georgia Tech. Implemented LoL-RL (Learning-over-Learning Reinforcement Learning) technique with co-design studies. Improved job placement rates by 35% across 200+ participants in pilot study.

GPT-4RLHFLoL-RLCo-designPersonalizationJob CoachingUser Studies
Narrative AI
NLP • Sci-Fi • Cornell

AI for Science Fiction Narrative Analysis

Developing NLP models for automated science fiction story structure analysis at Cornell Sci-Fi Lab. Built generative models for narrative pattern recognition and story arc prediction. Processing 10K+ sci-fi texts with transformer-based architecture achieving 78% accuracy on story structure classification.

NLPTransformersNarrative AnalysisGenerative ModelsStory StructureText ClassificationCornell

Application Development

Emotion AI
Biosensors • Real-time • iOS

EmBODY Real-time Emotion Classification

Built iOS app with Arduino biosensor integration for real-time emotion and pain tracking. Achieved 85% accuracy on emotion classification using custom CNN architecture. Deployed for 500+ patients, reducing manual assessment time by 60% and enabling immediate clinical interventions.

SwiftArduinoiOSCNNBiosensorsReal-time MLHealthcareClinical Deployment