PDF
Hamza Mohd Zubair

Hamza Mohd Zubair

Data Scientist & AI Engineer

Data scientist with 9+ years of deploying ML in production across consumer healthcare, retail, BFSI and QSR. Productized demand forecasting, supervised learning, reinforcement learning, deep learning, NLP, computer vision, Markov models and LLM pipelines. Also builds independent AI products and speaks on AI.

Experience

2026

EPAM Systems

Senior Data Scientist

Client: Opella (Sanofi Consumer Healthcare). Weekly e-commerce sell-out forecasting for Amazon and other retailers in 8+ markets, on Snowflake, SageMaker Pipelines and MLflow with Chronos-2, Prophet, MFLES and Ridge models.

  • Product discontinuation lifecycle in the production forecaster: phased-out products fade to zero, are capped by retailer stock and hand volume over to their successor.
  • Model vs. covariate study over 27 rolling backtest origins: showed model choice is worth 6–9 WAPE points while external data adds under 1, and ensembling gains about 2.
  • Plackett-Burman factorial screen of feature groups in 16 parallel SageMaker runs, replacing one-feature-at-a-time testing; cut screening time by 63% (246 → 91 min) and SHAP cost by about 60%.
  • Redesigned cannibalization detection to remove false positives from zero-inflated sales; found by reproducing runs that production had silently switched to the wrong Chronos checkpoint.
  • Audited the 9 forecast-exclusion rules (12.7% of recent volume excluded, several rules dead) and proposed a three-tier redesign; ranked cross-market data-quality issues by revenue impact.
  • Self-initiated internal tools: nx, a Snowflake data-access and run-reproduction library used in 40+ notebooks; benchpress, a FastAPI app for benchmarking forecast runs; and ops, a Rust CLI tying Jira, GitHub and SageMaker test runs together.
2024

Turing.com

Senior Data Scientist

  • Part of the team training Gemini with supervised fine-tuning (SFT).
  • Complex chart production tool (sole designer and developer): LLM agent pipelines that cut chart generation time by 93%, from 45 min to 3 min, at about ₹0.4 per chart.
  • Built RAG with optimized inference and caching, and production agentic workflows in LangChain and DSPy with observability and evaluation built in.
2018

Cartesian Consulting

3 promotions

Senior Lead Data Scientist

  • Campaign prioritizer (solo): contextual bandits (Thompson sampling, UCB) that pick each customer's campaign; deployed for 2 major clients (pharma retail, brokerage), lifting click-through and revenue.
  • Automated MLOps platform (team of 2): configure, deploy and maintain XGBoost / LightGBM models with automated feature engineering, tuning, calibration and retraining. Used by nearly all Solus clients.

Lead Data Scientist

  • SMS composer (solo): generative NLP module and UI that breaks a message down by purpose, tone and personalization fields and writes multiple variants from a prompt.
  • Multi-touch attribution (solo): Markov-model attribution adapted from digital to CRM and direct marketing.

Senior Data Scientist

  • Open rate predictor (led a team of 3): NLP pipeline that predicts email open rates from subject lines, plus a semantically controlled LSTM that writes subject lines. Deployed for a major Indian bank.

Senior Analyst – AI

  • Product segmentation (solo): LDA topic modelling to auto-categorize products for targeted marketing.
2017

Untrodden Labs

Founding Employee & AI Lead

  • Bootstrapped startup building an Alexa-like desktop robot with voice commands, vision, face tracking and an animated emotive display.
  • Led a team of 5 building speech intent classification (SVM, random forests), novelty detection (one-class SVM), information extraction (grammar and dependency parsing, SyntaxNet) and live object detection over 90 classes (YOLO, MobileNet SSD, ResNet v2).

Senior Data Scientist · EPAM Systems

Client: Opella (Sanofi Consumer Healthcare). Weekly e-commerce sell-out forecasting for Amazon and other retailers in 8+ markets.

  • Product discontinuation lifecycle in the production forecaster: phased-out products fade to zero, are capped by retailer stock and hand volume over to their successor.
  • Model vs. covariate study over 27 rolling backtest origins: showed model choice is worth 6–9 WAPE points while external data adds under 1, and ensembling gains about 2.
  • Plackett-Burman factorial screen of feature groups in 16 parallel SageMaker runs, replacing one-feature-at-a-time testing; cut screening time by 63% (246 → 91 min) and SHAP cost by about 60%.

Senior Data Scientist · Turing.com

  • Complex chart production tool (sole designer and developer): LLM agent pipelines that cut chart generation time by 93%, from 45 min to 3 min, at about ₹0.4 per chart.

Senior Lead Data Scientist · Cartesian Consulting

Promoted 3 times: Senior Analyst – AI → Senior Data Scientist (2019) → Lead (2020) → Senior Lead (2022).

  • Campaign prioritizer (solo): contextual bandits (Thompson sampling, UCB) that pick each customer's campaign; deployed for 2 major clients (pharma retail, brokerage), lifting click-through and revenue.
  • Automated MLOps platform (team of 2): configure, deploy and maintain XGBoost / LightGBM models with automated feature engineering, tuning, calibration and retraining. Used by nearly all Solus clients.
  • Open rate predictor (led a team of 3): NLP pipeline that predicts email open rates from subject lines, plus a semantically controlled LSTM that writes subject lines. Deployed for a major Indian bank.

Founding Employee & AI Lead · Untrodden Labs

Products

Redeo · Android, on Google Play

play.google.com

Spaced-repetition flashcards where you write the answer from memory, an on-device model scores your recall and FSRS schedules the next review. Kotlin, on-device ML, Room.

dontwatch.in · Web app

dontwatch.in

Learn a university lecture series by answering questions linked to the exact minute of each lecture; AI grades answers against what the lecturer said. Cloudflare Workers, LLM grading.

Education

Skills

Languages
Python, Rust, Kotlin, SQL, R
ML
PyTorch, TensorFlow, Keras, scikit-learn, XGBoost, LightGBM, SHAP, Vowpal Wabbit
Forecasting
Chronos-2, Prophet, MFLES, rolling backtests, factorial experiments
Data & MLOps
pandas, NumPy, SciPy, Snowflake, AWS SageMaker, MLflow, FastAPI, GitHub Actions
LLM
LangChain, DSPy, RAG pipelines
Areas
LLMs, agentic AI, NLP, bandits & RL, computer vision, forecasting, attribution

Awards

Talks