Professional Experience

• Leading a team of 20+ engineers building Generative Personalization and Foundation Models
• LLMs/SLMs Pre-training, Post-training (fine-tuning, preference alignment)
• Generative Personalization and Foundation Modeling for Atlassian Intelligence

• Udacity: Advanced LLM Inference Optimization Techniques — covers transformer tweaks, model parallelism, and sharding using DeepSpeed, TensorRT-LLM, and Triton Inference Server
• LinkedIn Learning: Building LLM-Powered Recommendation Systems — production-ready recommender systems using GenAI for enhanced personalization

• Founded and led the Personalization ML team (0→1); oversaw 15+ engineers across 5+ products
• Built and deployed Deep Learning models from scratch for conversational commerce (text/email marketing) across hundreds of millions of end-users and 1,000+ brands
• Advanced neural architectures: MLP stack, deep & cross networks, residual connections, attention, MMoE, PLE
• Distributed model training, mixed precision training, ANN vector store-based serving
• Published engineering blog on personalization work (Oct 2024)
• Technical leadership in defining long-term personalization strategy, partnering with senior executives

Tech: Python, PyTorch, Spark, Airflow, Feature Store, ANN Serving, A/B Experimentation

• Invited Research Committee Member at 20+ leading AI conferences
• Conferences: TheWebConf (2x), CIKM (3x), ICML (1x), SIGKDD (2x), SIGIR (3x), ICDM (2x), RecSys (3x), ECML-PKDD (3x), ICWSM (3x), and TORS (1x)
• Published author in LLMs (NLP and GenAI) and Recommender Systems domains, amassing 1,150+ citations

• Founding engineer of the Conversations Quality team applying ML and NLP to drive healthy conversations from Reply product; also worked on Home Feed Ranking
• Built Deep Learning models in TensorFlow (shallow → deep → multi-task learning), driving >30% gains in key engagement metrics
• Architected KubeFlow pipelines on GCP (DataFlow + BigQuery) to improve model training duration by 10x
• Shipped graph embeddings (Heterogeneous Information Network) and text embeddings (fine-tuned BERT), encoding user preferences and author affinity
• Developed ML-powered explainable reply ranking feature — improved reply health and follows by 3–5%
• Architected low-latency Early Ranking system handling tens of millions of candidates per second; 20% improvement in p9999 latency while improving health metrics by 5%
• Patent filed based on reply ranking work
• Built roadmaps via data-backed analyses & shipped measurement frameworks

Tech: Python, Java, Scala, TensorFlow, Airflow, KubeFlow, BigQuery, DataFlow, GCP, Scalding

• Organized and led technology and AI-focused events, workshops, and meetups

• Infrastructure scalability and ML applications for Amazon's international shipping business (Expansions & Exports)
• Improved infrastructure using native AWS technologies; improved eligibility prediction using ML models
• Teaching Assistant for Amazon's ML University (Intro to Data Science, Text Mining)
• Reviewer for Amazon's Machine Learning Conference (AMLC) 2019
• Presented product size recommendation research internally and at AMLC

Tech: AWS (Lambda, SQS, SNS, SageMaker, CloudWatch, DynamoDB), Java, Python

• TA for Recommender Systems & Web Mining (CSE 258, Prof. Julian McAuley) and Software Engineering (CSE 110, Prof. William Griswold)
• Led labs on Android development and project management

• Led development of distributed platform for efficient scheduling of Big Data workloads in the DataForge team
• Implemented primary key constraints, batch inserts/updates, and transactionality in Hive

Tech: Java, Hive, DynamoDB

• Led low-latency processing of critical financial artifacts in the STP team
• Migration from legacy to Java-based infrastructure with self-sanitization, self-recovery, and fault tolerance
• Profiled and optimized code (~40% improvement using concurrency) and database (indexes + partitions)

Tech: Java, Spring, MyBatis, SQL Server

• Developed scalable Bayesian Matrix Factorization (cubic → linear complexity) under Dr. Balaraman Ravindran
• Developed scalable variational Bayesian framework for Factorization Machines

Tech: C++, Python, MATLAB

• Specialization in Machine Learning, Recommender Systems, and NLP
• Advisor: Prof. Julian McAuley