Vinayak Bassi
Atlanta
Summary
Applied Scientist with 5+ years of experience building scalable ML and optimization systems for personalization and pricing. Expertise in reinforcement learning, causal inference, and LLM systems, driving impact across revenue and customer engagement.
Overview
7
7
years of professional experience
Work History
Applied Data Scientist
The Home Depot
Atlanta
05.2025 - Current
- Built an off-policy RL next-best-product optimizer on customer time-series data (TensorFlow, distributed GPU), achieving 85% top-3 DD recall; deployed via Vertex AI pipelines; patent pending
- Developed Transformer-based propensity models, improving top-1 DD recall by ~30% (OOT); optimized audience selection via Lagrangian post-scoring, driving ~12% incremental revenue lift
- Engineered sparse time-series feature pipelines across 50M+ households (across 52 weeks horizon), reducing training time by ~35%
- Improved enterprise NL2SQL alignment using GRPO, achieving ~3-4% precision gain; trained an SLM via on-policy distillation
- Designed and executed A/B tests, delivering +10 bps conversion and +7 bps CTR lift
Graduate Student Research Assistant
University of Michigan
Ann Arbor
01.2024 - 04.2025
- Developed RL-GNN models for influence maximization on temporal COVID spread networks, leveraging link prediction and reward shaping to outperform heuristic baselines; co-authoring manuscript
- Identified causal confounders from clinical text (DAG-based modeling) for opioid prediction (F1 ~0.84); used Shapley analysis for bias detection
- Improved robustness via counterfactual text generation and OOD generalization in LLMs, achieving ~6-8% accuracy gains under distribution shift
Data Science Intern
Tesla
Fremont
05.2023 - 08.2023
- Reduced transportation costs by ~30% using MILP (Gurobi), with additional ~8% reduction via gradient-boosted demand forecasting
- Applied RLHF to fine-tune a 770M T5 model, improving precision by ~20% for inventory overflow classification
- Optimized inference (FP16), reducing latency by ~25%; deployed via Kubernetes, achieving 98% UAT pass rate
Data Scientist
Amazon
06.2020 - 07.2022
- Developed Bayesian PED models (~7.5% MAPE) to optimize pricing decisions, enabling advanced shipping carrier selection with ~8% sales attrition on higher ASP segments
- Applied causal inference (Propensity Score Matching, Difference-in-Differences), driving ~12% retention lift and ~15% sales growth
- Built GNN models (GCN, GraphSAGE), improving delivery accuracy by ~18%
- Designed A/B-tested marketing campaigns, increasing registrations by ~38% and improving ROI by ~15%
Associate Data Scientist
United Airlines
07.2019 - 06.2020
- Deployed aircraft maintenance risk models (Palantir), reducing workload deferrals by ~55% and improving operational efficiency
- Optimized PySpark pipelines, reducing compute time by ~80% and improving uptime by ~20%
Education
MS - Computer Science & Engineering
University of Michigan
Ann Arbor, MI04-2025
MS - Industrial & Operations Engineering
University of Michigan
Ann Arbor, MI04-2025
B.Tech - Aerospace Engineering
Punjab Engineering College
India04-2019
Skills
- Programming: Python, SQL, C
- Frameworks: PyTorch, TensorFlow
- Cloud: GCP, AWS, Palantir
- MLOps: Docker, Kubernetes, ADK
Publications
- Reinforcement learning-based next-best-product optimization system for marketing, patent filing in progress
- Numerical Study of Deflagration to Detonation Transition using OpenFOAM (C++), IEEE Aerospace Conference (2020)
- Numerical Study of Deflagration to Detonation Transition using OpenFOAM (C++), Springer (2019)
Timeline
Applied Data Scientist
The Home Depot
05.2025 - Current
Graduate Student Research Assistant
University of Michigan
01.2024 - 04.2025
Data Science Intern
Tesla
05.2023 - 08.2023
Data Scientist
Amazon
06.2020 - 07.2022
Associate Data Scientist
United Airlines
07.2019 - 06.2020
MS - Computer Science & Engineering
University of Michigan
MS - Industrial & Operations Engineering
University of Michigan
B.Tech - Aerospace Engineering
Punjab Engineering College