Projects
11 projects spanning data engineering, machine learning, cloud infrastructure, and full-stack development - built at IBM, Samsung, UCSD, and KL University.
GridGreen
A developer-first copilot that analyzes ML training code before execution - estimating CO₂, suggesting greener model alternatives, and scheduling jobs in cleaner grid windows.
GPU Kernel + Distributed Training
Fused Triton GPU kernel (D=ReLU(A×B+C)) achieving ≥1.25× PyTorch speedup via shared memory tiling and operator fusion, plus 2D data × tensor parallel transformer training with custom MPI primitives.
AutoDiff Engine + Transformer LM
Custom automatic differentiation engine and a decoder-only Transformer language model built from scratch - no PyTorch autograd. Full computational graph, 8 operators, causal attention, and autoregressive generation.
Large-Scale NYC Taxi Pipeline
Distributed analytics pipeline processing 57GB+ NYC taxi data with Dask and PyArrow - PCA-based representation learning, heavy-tail statistical analysis, bootstrap stability, GAM fare prediction, and XGBoost classification.
MediDB – Drug Safety & Recommendation
Multi-database clinical decision-support system combining SQL, graph, and vector retrieval - explainable drug safety assessments with DDI detection, a Streamlit UI, and Docker deployment.
Predicting House Prices in Taipei
Rigorous regression analysis of 414 Taipei transactions - LASSO, 10-fold CV (best CV MSE 67.2, R² 0.71 on log scale), bootstrap CIs 28-30% wider than classical, and logistic model with 82.6% accuracy, AUC 0.92.
Customer Response Analysis
Analysed 40,000+ bank marketing records with Python and Tableau - surfacing behavioral segments (retired 34%, students 28% conversion) and channel patterns that improved targeting efficiency by 12%.
Socially-Aware Recommendation System
Production-ready recommendation system with Bayesian MRF and social trust networks on 664K+ Epinions reviews - AUC 0.6248, 25% over baseline, 6% over IJCAI 2017, deployed on Hugging Face with FastAPI.
Multi-Cloud Provisioner
LLM-powered provisioner built at IBM - natural language to optimised IBM Cloud resource configs, cutting costs by 40% and validation time by 90% in production. Star of the Month award.
Eco Entrepreneurship
Full-stack platform promoting green business in India - startup idea recommendations, environmental impact calculators, and a marketplace deployed on AWS EC2 + RDS + S3.
Amazon Web Services Project
Production-grade AWS hosting - EC2 with Apache, VPC network isolation, Elastic Load Balancer, Auto Scaling groups, and CloudWatch monitoring.
Online Survey System
Survey platform with Java Servlets and JSP - customisable templates, shareable links, real-time response dashboards with Chart.js, and secure user data management.
Phoneme Data Creation
Multilingual phonetic dataset of 10,000+ names with IPA annotations - Soundex/Metaphone validation achieves 10% accuracy improvement in speech recognition at Samsung Prism.
Take A Trip
Award-winning travel planning platform - Dijkstra route optimisation (up to 20% cost savings), hotel booking, agent connections, and a Figma-designed responsive interface.
No projects match
Try adjusting your filters or search term.