Projects & Production ML
A selection of production ML systems, research projects, and data science work spanning e-commerce, construction tech, luxury resale, and NLP.
Luxury Resale Pricing & Demand Forecasting
Leading ML-powered pricing algorithms and inventory forecasting at Fashionphile, one of the largest luxury resale platforms in the US. Owns the data infrastructure powering real-time pricing decisions across thousands of SKUs.
2025 - Present
Graph-Based E-commerce Recommendation Engine
Developed a graph-based recommendation system using DeepWalk, Node2Vec, and FAISS for efficient similarity search on e-commerce clickstream data, enhancing product recommendations.
Nov 2023 - Feb 2024
Real-Time Fraud Detection — $12M Savings
Developed real-time fraud detection using ensemble methods (Random Forest, XGBoost, neural nets), cutting fraud by 60% and saving $12M yearly at a NASDAQ-listed e-commerce company.
2021 - 2024
LSTM Demand Forecasting & Inventory Optimization
Orchestrated LSTM forecasting and GCP pipelines (BigQuery, Dataflow), slashing inventory costs by 15% (~$8M) and stockouts by 35%. Designed inventory optimization algorithms improving fulfillment efficiency by 22%.
2021 - 2024
Customer Segmentation & A/B Testing Platform
Led 100+ A/B tests improving conversion by 7% (~$15M incremental sales). Built customer segmentation with Spark ML, identifying segments generating 45% of revenue and elevating retention by 25%.
2021 - 2024
BERT NLP Pipeline for Document Classification
Deployed BERT-based NLP pipelines on 40K+ internal documents for topic modeling and classification. Reduced project scoping time from 2 weeks to 2 days. Built Python Flask frontend for end-to-end ML POC.
2024 - 2025
ML Cost Estimation Models — 95% Accuracy
Engineered a Random Forest Regression model achieving 95% accuracy in forecasting FTE costs. Deployed Azure web apps with CI/CD pipelines enabling instant cost predictions for project managers.
2024 - 2025
Employee Sentiment Analysis Pipeline
Built a multi-stage NLP pipeline for survey data (10,000+ responses) with 92% classification accuracy and 88% F1-score in sentiment analysis, directly informing engagement policies for 5,000+ employees.
2024 - 2025
Attention-Enhanced Text Classification
Developed an attention-based neural network in PyTorch for multi-class classification of 2M+ financial consumer complaints, improving accuracy over baseline NLP models.
Nov 2023 - Feb 2024
American Sign Language Recognition
Research comparing deep learning models (CNN, VGG16, Inception Net, RESNET) for ASL gesture recognition. Published paper on Semantic Scholar.
2019 - 2020
RNNs for Visual Question Answering
Improved VQA systems by integrating modern text-processing techniques and experimenting with combinations of high-performing neural network architectures (RNNs, GRUs).
2020
Airbnb Price & Superhost Prediction
Used advanced ML techniques to predict Airbnb listing prices and identify Superhosts. Achieved 91% accuracy on test data.
2019
Gesture Genie
Integrated Magenta's Piano Genie with Google's Hand Tracker to create a deep learning based, user-friendly musical tool for generating music through hand gestures.
2020
ML Characterization in Low Power Embedded Devices
Developed machine learning programs for low-power embedded devices such as Arduino and Raspberry Pi 3, testing their processing capabilities for ML applications.
2019
Text Classification and Sentiment Analyzer
Developed a system capable of classifying user input into distinct categories and performing in-depth sentiment analysis, using the LIME architecture for model explanations.
2019
Portable Navigation System for Visually Impaired
Developed Python scripts to create a portable navigation system for visually impaired individuals during an internship at Bangkok University Robotics Lab.
2017