Gaurav Kumar

Executive Summary

Accomplished Machine Learning Scientist with 6+ years of experience designing, implementing, and deploying high-impact AI/ML solutions in production environments. Since graduation from IIT Kharagpur (2019), maintained unwavering commitment to understanding machine learning from first principles—building recommendation systems, deep learning architectures, and optimization algorithms from scratch.

Unique Value: I don’t just use frameworks; I implement them. This from-scratch approach enables production optimization, innovative solutions, and deep mentorship that others cannot achieve.

Key Expertise: Recommendation Systems (TARNET, DeepFM, DCN v2), Deep Learning, Optimization Theory, Large-Scale ML Systems

Core Competencies

Recommendation Systems & Architecture Implementation

• Frameworks Built From Scratch: TARNET (Two-Arm ResNets), DeepFM (Deep Factorization Machines), DCN v2 (Deep & Cross Networks), Wide & Deep Networks, Multi-task Learning Architectures, Ranking Networks
• Understanding: Embedding optimization, feature cross mechanisms, gradient flow analysis, inference optimization
• Impact: 225% CTR improvement, 3% attach rate increase, 2.3% margin growth

Machine Learning & Optimization

• Gradient Boosting: XGBoost internals, tree-based learning, split optimization, pruning strategies
• Advanced Optimization: Hessian-based methods, second-order optimization, Newton methods, convergence analysis
• Algorithmic Techniques: Weighted quantile sketching, dimensionality reduction (PCA), Bayesian inference, game theory

Deep Learning & NLP

• Architectures: Transformer models, BERT, attention mechanisms, encoder-decoder systems, sequence-to-sequence models
• Techniques: Transfer learning, entity extraction, word embeddings, language model fine-tuning
• Production: Large-scale inference optimization, model serving, quantization

ML Engineering & Systems

• Scale: Systems handling millions of users, distributed computing, real-time recommendations
• Optimization: Feature engineering pipelines, data processing with PySpark, MLOps, CI/CD deployment
• Infrastructure: AWS (EC2, Bedrock), Databricks, Docker, feature stores

Professional Experience

Expedia Group

Machine Learning Scientist | July 2024 – Present

Search, Ranking, Recommendations, Gen-AI Automation

• Search & Ranking: Improved NDCG@10 by 6% and Global Profit Margin by 1.2% through advanced ranking algorithms and custom optimization
• TARNET Recommendation Engine: Built Two-Arm ResNet architecture from scratch for personalized travel insurance recommendations. Results: +3% attach rate, +2.3% Gross Profit Margin
• Custom Neural Architectures: Leveraged transfer learning combined with custom deep learning implementations to achieve state-of-the-art performance
• Gen-AI Automation: Engineered Generative AI solution automating B2B partner supply complaint resolution. Projected annual savings: 13% of support operations budget
• Technical Leadership: Architected inference pipelines, optimized model serving, mentored team on from-scratch implementation practices

Nykaa

ML Engineer II | July 2023 – July 2024

Recommendation Engines, CTR Optimization, Feature Engineering

• Deep Learning Recommendations: Designed and implemented recommendation systems suggesting unexplored categories based on behavioral signals (views, orders, wishlist)
• CTR Optimization: Improved CTR by 225% versus traditional banners and 20% above platform average through feature interaction models
• Feature Engineering: Enhanced “More from Category” and “More from Brand” algorithms by 50% through mathematical optimization
• Multi-objective Learning: Built architectures optimizing simultaneously for engagement, conversion, and user satisfaction
• Scale & Performance: Deployed systems handling millions of daily users with sub-100ms inference latency

OLX Autos

Data Scientist | July 2022 – July 2023

Pricing Models, Ranking Algorithms, Large-Scale Optimization

• ML-Based Pricing Models: Developed Web Quote Pricing for LATAM (Chile & Mexico). Live on 80% and 20% traffic respectively
• Custom Ranking Algorithm: Designed ranking system for car model trims incorporating market dynamics, user preferences, and inventory signals
• Optimization Results: Reduced sell price prediction error to <3% and increased per-car margin by >30% in Chile
• Data Pipeline: Built PySpark feature pipelines processing millions of vehicle records for large-scale predictions

Publicis Sapient

Senior Associate, Data Science | July 2019 – July 2022

Computer Vision, NLP, Optimization, Price Analytics

• Computer Vision Pipeline: Developed YOLO-based document sectioning algorithm for resume mapping and understanding
• NLP & Entity Extraction: Built transfer learning-based NER model extracting Name, College, Degree, Company, Skills with high accuracy
• Optimization Systems: Developed price optimization using K-means clustering and mathematical optimization on user purchase behavior
• Media Analysis: Extracted and analyzed objects and text from creatives to measure impact on CTR performance
• Foundation: Established from-scratch implementation philosophy in first role out of college

Education

Indian Institute of Technology (IIT), Kharagpur
Bachelor of Technology in Chemical Engineering | 2015 – 2019

Technical Skills

Programming Languages: Python, SQL, LaTeX, Markdown

ML & Deep Learning Frameworks: PyTorch, TensorFlow, XGBoost, Scikit-learn

Specializations:

• Recommendation Systems: TARNET, DeepFM, DCN v2, Wide & Deep, Multi-task Networks
• Optimization: Embedding learning, feature crosses, attention mechanisms, Hessian-based methods
• NLP: BERT, Transformers, embeddings, entity extraction, language models
• Gradient Boosting: XGBoost internals, tree optimization, gradient descent variants
• Advanced ML: Bayesian inference, game theory, dynamic programming, probabilistic models

Cloud & DevOps: AWS (EC2, Bedrock), Databricks, Docker, CI/CD, Git, MLOps

Data Technologies: PySpark, Jupyter, Feature Stores, Distributed Computing

Key Achievements

Professional Philosophy

From-Scratch Implementation (Since 2019)

My career is built on one principle: understand architecture by building it. I don’t use TARNET, DeepFM, or DCN—I implement them from scratch multiple times, understanding every mathematical detail, gradient computation, and optimization opportunity. This depth enables production optimization that practitioners relying on high-level APIs cannot achieve.

Mathematical Rigor

Every model is grounded in mathematics. I understand feature crosses, embedding interactions, attention mechanisms, and loss functions at fundamental levels. This rigor compounds over 6 years, enabling systematic optimization and innovation.

Production Excellence

I bridge theory and practice. Systems I build scale to millions of users and deliver measurable business impact: 225% CTR improvement, <3% pricing error, 6% ranking improvement, 13% cost savings.

Continuous Optimization

Each challenge receives systematic mathematical analysis and optimization. Whether improving CTR by 225%, reducing error to <3%, or building architectures, I approach problems with deep expertise and methodical refinement.

Educational Leadership

I mentor through code. Repositories and implementations serve both as production systems and learning resources, demonstrating how to build scalable, well-designed systems.

Consistency Over Time

From Publicis Sapient (2019) through Expedia (2024), maintained same core philosophy: understand by building, optimize through mathematics, ship systems delivering business impact.

Current Focus & Research Interests

Recommendation Systems Evolution

• Advanced architectures (DCN variants, novel networks)
• Large-scale embedding optimization
• Multi-objective learning for business metrics
• Real-time adaptation and online learning
• Inference optimization and quantization

Optimization & Mathematical Foundations

• Hessian-based and second-order optimization methods
• Weighted quantile sketching for gradient boosting
• Feature interaction theory and mechanisms
• Probabilistic inference and Bayesian optimization
• Game theory applications in ML

Emerging Technologies

• Transformer mechanics and attention systems
• Reinforcement learning for recommendations
• Knowledge distillation and model compression
• Continual learning in production systems
• Large language model optimization

Let’s Connect

I’m open to collaborating on recommendation system research, discussing ML architecture design, exploring research partnerships, and mentoring engineers passionate about building systems from first principles.

Email: gauravkr927@gmail.com
LinkedIn: linkedin.com/in/gauravkr8
GitHub: github.com/Gaurav927
Portfolio: gaurav927.github.io

Last Updated: December 2025

6+ years of commitment to understanding machine learning from first principles and building systems that scale.