⬢ Machine Learning Scientist  ·  Ph.D.

Mithun Ghosh

Meta  ·  Walmart Labs  ·  Deep Learning · LLMs · Scalable ML Systems

Building production ML at scale — from Transformer-based forecasting and Foundation Model fine-tuning to recommendation systems driving revenue across billions of ad auctions.

0
Years Industry Exp.
0%
Inference Time ↓
0%
MAPE Reduced
0%
Revenue Lifted
0%
Model Accuracy
0
Publications
Scroll

ML Scientist.
Industry & Research.

I'm a Machine Learning Scientist with a Ph.D. from the University of Arizona, with 6+ years of industry experience at Meta and Walmart Labs building and shipping large-scale ML systems.

My work spans deep learning model development, Foundation Model fine-tuning, Transformer-based time series forecasting, auction-based recommendation systems, and operational ML — handling billions of data points daily in production.

I bridge the gap between cutting-edge research and real business impact: cutting inference costs, lifting revenue, and accelerating model iteration across some of the world's largest technology companies.

Deep Learning Foundation Models Time Series RecSys Bayesian ML Optimization Computer Vision NLP / BERT Anomaly Detection
XGBoost — Proposal Stage Prediction85%
Bayesian GP — Anomaly Detection92%
Transformer — MAPE Reduction30%
Few-Shot LLM — Inference Speedup60%
DeepFM — Revenue Uplift15%
AI Tool Productivity Lift25%

Tools & Technologies

A full-spectrum ML toolkit — from model development to cloud deployment and big-data pipelines.

🐍

Languages

PythonR JavaSQL
🧠

ML / Deep Learning

PyTorchTensorFlow XGBoostscikit-learn DeepFMCNNs
🤖

LLMs & Foundation Models

TransformersBERT TinyTimeMixtureLag-Llama Few-ShotFine-Tuning
☁️

Cloud & Infra

GCPAWS BigQueryVertex AI

Big Data & Databases

PySparkHive MySQLMSSQL BigQueryPower BI
📊

Domains

Time SeriesRecSys Computer VisionNLP OptimizationBayesian Anomaly Detection

Professional Experience

Production ML systems serving millions of users at FAANG and top-tier tech.

Meta
Senior Data Scientist
Feb 2025 – Present
📍 Menlo Park, CA
  • Forecasted adoption metrics for AI tools using ML and statistical models, driving productivity up by +25%.
  • Built an XGBoost model to predict proposal stage progression to production with 85% accuracy.
  • Designed methodology for ML model ROI estimation and operational cost savings to set and track ROI progress across AI initiatives.
Walmart Labs
Senior Data Scientist
May 2022 – Feb 2025
📍 Sunnyvale, CA
  • Reduced MAPE by 30% using Transformer-based models for time series demand forecasting.
  • Applied few-shot learning on Foundation Models (TinyTimeMixture, Lag-Llama), cutting inference time by 60%.
  • Improved revenue by +15% by deploying DeepFM for auction-based recommendation systems.
  • Built and deployed a dynamic floor pricing optimization framework, boosting ad display by +5%.
Lumen Technologies
Data Science Intern
Jun – Aug 2021
📍 Broomfield, CO
  • Developed scalable forecasting tools to optimize headcount allocation across business units.
  • Automated SQL + Power BI pipelines for workforce analytics and product line reporting.
University of Arizona
Graduate Research Assistant
Aug 2017 – May 2022
📍 Tucson, AZ
  • Built Bayesian Gaussian Process model for anomaly detection achieving 92% accuracy.
  • Applied CNNs for X-ray image classification, accelerating medical image triage.
  • Developed NLP-based sentiment analysis model with +9% improvement over SOTA benchmarks.

Featured Projects

Applied ML projects spanning computer vision, NLP, time series, and recommendation systems.

Deep LearningComputer Vision

Ensemble CNN for MURA X-Ray Classification

Ensemble of CNN architectures to classify abnormal vs. normal X-rays from the MURA musculoskeletal dataset. Different CNN backbones detect diverse fracture types, boosting ensemble accuracy over single models.

View on GitHub
NLPBERT

Sentiment Analysis of Tweet Data Using BERT

Multi-class sentiment analysis of tweet data using fine-tuned large-case BERT. Outperforms RNN, LSTM, and GRU baselines, demonstrating the power of Transformer-based pre-training for NLP tasks.

View on GitHub
Time SeriesForecasting

Time Series Forecasting of Panel Data

Panel data forecasting with cross-sectional and temporal variation. Predictive models for user mission-completion in a health app — combining time-series and cross-sectional analysis techniques.

View on GitHub
NLPDeep Learning

Article Popularity Prediction via NLP

Deep learning model using embedding projectors to predict article upvote/downvote status from title text. Includes Tableau visualizations to surface key dataset characteristics.

View on GitHub
MLClassification

Imbalanced Class Classification via Ensembling

Salary quartile classification on highly imbalanced data. Explored SMOTE, up/downsampling, and ensemble ML models. Evaluated via F1 and ROC-AUC to select the best-performing approach.

View on GitHub
Fraud DetectionML

Credit Card Fraud Detection

ML solution for detecting fraudulent credit card transactions on a highly imbalanced dataset (0.17% fraud rate). Handles class imbalance with appropriate evaluation metrics and model selection.

View on GitHub

Publications

Peer-reviewed research in applied ML, Bayesian methods, and industrial engineering.

Academic Background

🎓
Doctor of Philosophy
Systems & Industrial Engineering
The University of Arizona · Tucson, AZ
May 2022
🏛️
Bachelor of Engineering
Industrial & Production Engineering
Bangladesh University of Engineering & Technology
September 2015

Get In Touch

Open to senior MLE roles, applied ML research collaborations, and speaking on ML at scale.