Debanjan Saha

Summary

Seasoned Machine Learning Engineer with 10+ years of experience architecting Generative AI, LLM, and multimodal ML systems across multi-cloud environments (AWS, Azure, GCP). Adept at end-to-end ML lifecycle management, model optimization, and real-time AI deployment. Proven leader in cross-domain projects spanning marketing analytics, financial risk modeling, and healthcare AI. Seeking Lead Machine Learning Engineer role advancing foundation model development and scalable AI infrastructure.

Skills

ML Frameworks

PyTorch, TensorFlow, Keras, Scikit-learn, Spark ML, Hugging Face, OpenAI, LLM, RAG, LangChain, LlamaIndex, MCP

DevOps & MLOps

Docker, Kubernetes, AKS, EKS, GKE, Helm, Terraform, MLflow, Kubeflow, ELK, Prometheus, Databricks

Cloud Ecosystems

AWS (SageMaker, S3, EMR), Microsoft Azure (Azure ML, ADLS, AI Search), GCP (Vertex AI, Cloud Run)

Programming & Databases

Python, Scala, PySpark, Matlab, C/C++, SQL (PostgreSQL, MS-SQL, MySQL), HTML/CSS, React, Node.js

Data Visualization & Analytics

Tableau, PowerBI, QuickSight, Grafana, Looker, Jupyter Notebook, Data Storytelling

Experience

1. Mastercard

   Senior Data Engineer

   January 2025 - Present

   Dublin, Ireland

   • Engineered large-scale data pipelines processing ~45 TB of daily transaction data across Mastercard systems, enabling advanced analytics use cases including segmentation, lifetime value, churn, affinity, loyalty, and share-of-wallet modeling.
   • Contributed to the development of Pioneer, an internal privacy-enhanced analytics platform powering 200+ Mastercard users, providing real-time access to curated datasets with a Zeppelin-based notebook interface.
   • Architected secure, high-performance data access layers replacing legacy Hadoop workflows, reducing data retrieval latency by >70% and improving query reliability.
   • Prototyped Text-to-SQL capabilities into Pioneer using LLMs, enabling natural language query generation to democratize data access.
   • Designed and maintained ML pipelines in Azure Databricks and MLflow, supporting end-to-end model lifecycle management.
   • Ensured compliance with PCI-DSS and GDPR standards while enabling secure feature engineering and model deployment at scale.
   • Advocated internal ML infrastructure evolution, driving automation using Terraform, reproducibility, and cost optimization via Photon compute, Delta Lake optimization, and parallelized data ingestion.

   • PySpark
   • Azure Databricks
   • Delta Lake
   • MLflow
   • Terraform
   • LLM
   • Text-to-SQL
   • PCI-DSS
   • GDPR

2. AT&T

   Machine Learning Engineer

   September 2024 - January 2025

   Dallas, TX, USA

   • Built an end-to-end multimodal AI pipeline to digitize over 10M+ legacy tower contract PDFs, leveraging Vision Transformers, PyMuPDF, and OCR models (Tesseract, OCRmyPDF), improving accuracy by 34% over baseline OCR.
   • Benchmarked and integrated state-of-the-art embedding models (OpenAI, Azure, Sentence-Transformers), increasing retrieval precision by 42% in Azure AI Search.
   • Architected an Azure-based knowledge retrieval system storing embeddings and document metadata in AI Search + ADLS, enabling instant semantic search and metadata-aware insights.
   • Fine-tuned a LLaMA-based Large Language Model on internal contract embeddings and metadata, reducing legal drafting time by ~60%.
   • Implemented query tokenization, reranking, and Retrieval-Augmented Generation (RAG) in production search UI, improving relevance and latency by >25%.
   • Deployed scalable inference endpoints on Azure Kubernetes Service (AKS) and integrated APIs into the ContractsAI web interface.

   • LLaMA
   • RAG
   • Vision Transformers
   • Azure AI Search
   • Azure Kubernetes Service (AKS)
   • PySpark
   • Azure Databricks
   • OpenAI Embeddings

3. Nanobiosym

   Principal AI Engineer

   May 2023 - December 2023

   Cambridge, MA, USA

   • Architected a cloud-agnostic Generative AI platform on Kubernetes (EKS) and Docker within a HIPAA-compliant AWS environment, delivered in under 6 months for multi-tenant scalability.
   • Developed LSTM-based COVID-19 severity prediction models leveraging patient vitals and longitudinal data, reducing intervention response time by 30%.
   • Fine-tuned domain-specific NER models on patient–clinician dialogues, achieving an F1 score of 0.87 in symptom extraction and severity assessment.
   • Deployed a conversational AI assistant using Amazon Lex and Polly, enhancing patient engagement by 40% through real-time symptom triage.
   • Implemented shadow binning and SNS-triggered alerting for critical patient classification, cutting emergency response times by 20%.

   • AWS (EKS, SageMaker, Lex, Polly, SNS)
   • Generative AI
   • LSTM
   • NER
   • Docker
   • HIPAA Compliance

4. Prescriber 360

   Senior Data Scientist

   September 2021 - July 2022

   India

   • Optimized attribution and marketing mix models (MMMs & MTAs) to evaluate channel effectiveness, improving ROI by 15–20% through regression-driven media allocation.
   • Enhanced user journey modeling and conversion tracking, increasing acquisition efficiency by 12% via predictive funnel insights.
   • Developed hybrid LSTM time-series models to forecast campaign performance and engagement.
   • Automated KPI monitoring pipelines for QPS, LTV, and funnel analysis using Spark and Delta Lake.
   • Led enterprise code review initiatives, reducing defects by 80% and standardizing development practices across teams.

   • PySpark
   • Delta Lake
   • LSTM Time-Series
   • Marketing Mix Modeling (MMM)
   • MTA
   • Attribution Models

5. Cognizant

   Senior Data Scientist

   October 2019 - September 2021

   India

   • Optimized two-tower recommendation architecture for personalized product search, increasing ranking precision and CTR by 18%.
   • Built scalable ML pipelines on AWS (SageMaker, Glue, Airflow, S3), cutting model deployment time by 40%.
   • Enhanced large-scale ranking pipelines on AWS EMR, reducing product search latency by 25% through scoring algorithm optimization and LSH acceleration.
   • Implemented semantic search models using Word2Vec and custom NLP embeddings, improving query relevance and diversity by 15–20%.
   • Developed fairness-aware post-processing layers to balance promoted product visibility while maintaining top-ranking relevance.

   • AWS (SageMaker, Glue, EMR, S3)
   • Airflow
   • Two-Tower Recommender
   • Semantic Search
   • Word2Vec
   • NLP

6. Tata Consultancy Services (TCS)

   Data Engineer

   December 2015 - September 2019

   India

   • Developed rule-based AML risk scoring modules using Hadoop and HiveQL, improving suspicious activity detection accuracy and reducing false positives by 30%.
   • Engineered end-to-end pipelines for feature extraction from behavioral, temporal, and network transaction data.
   • Migrated legacy SQL systems to cloud to enable scalable Anti-Money Laundering (AML) monitoring and faster SAR investigation workflows.

   • Hadoop
   • HiveQL
   • Big Data Pipelines
   • AML Risk Scoring
   • SQL Migration

7. ITC Infotech

   Associate Data Warehouse Engineer

   July 2013 - November 2015

   India

   • Optimized large-scale SQL workloads through schema redesign, materialized views, and indexing, improving query performance by 25–40% and saving 1,800+ analyst hours annually.
   • Implemented workload management and caching strategies to prioritize critical jobs and enhance resource utilization by 20%.

   • SQL Server
   • Data Warehousing
   • Query Optimization
   • Materialized Views
   • Performance Tuning

Research

Education

Certifications

• AWS Certified Solutions Architect – Associate

  Amazon Web Services

• AWS Certified Developer – Associate

  Amazon Web Services