Jul 2025 – Aug 2025 · 1 mo · Remote

• Built an explainable ML-based offer personalization engine that ranks promotions using user transaction behavior, location signals, merchant affinity, category preference, cashback sensitivity, and temporal patterns.
• Designed production-ready pipelines and APIs, including cold-start handling, confidence scoring, and match rationale, with seamless backend integration.
• Delivered a maintainable, model-agnostic codebase (training, inference, evaluation) handed off via GitHub for full internal ownership.

Jun 2024 – Jun 2025 · 1 yr · United States · Remote

• Selected among the Top 15 LLM Engineers to work and research on POCs and my project was recognized among the Top 3 which can also be seen in the company portal.
• APOLLO GLOBAL MANAGEMENT
• AI-Powered Email Assistant Chatbot for Outlook (Agentic RAG & Conversational AI)
• Built an advanced RAG-based Outlook assistant leveraging LangChain, LangGraph, FastAPI, and Azure services
• Used GPT and document retrieval with cosine similarity to generate accurate, context-aware responses
• Deployed using FastAPI on Azure Kubernetes with Cosmos DB for metadata storage
• Impact: Achieved \~3s latency, reduced manual drafting effort by 90%, and boosted enterprise productivity
• Travel Gear Recommendation Agent (Multi-Agent System + RAG)
• Designed a multi-agent RAG system using Gemini for personalized travel product recommendations
• Implemented PostgreSQL-backed conversation memory and user profiling
• Impact: Improved recommendation relevance by 40% and reduced development time by
• Domain-Specific Model Optimization
• Fine-tuned LLMs using structured instruction templates in medical, legal, and financial domains
• Impact: Boosted accuracy and compliance in sensitive, high-stakes environments
• GOOGLE(AI Research -LLM)
• LLM Training and Fine-Tuning (SFT + RLHF)
• Curated high-quality SFT datasets for Google Gemini RLHF pipeline
• Designed stress test prompts in domains like math, physics, and logical reasoning
• Impact: Achieved 78% response accuracy via structured fine-tuning and prompt engineering
• RLHF – Reinforcement Learning from Human Feedback
• Designed reward models, curated preference datasets, and trained models for safer, stylistically aligned LLM behavior
• Impact: Improved model alignment with user expectations and real-world applicability
• LLM Output Evaluation and Quality Assurance
• Assessed outputs for correctness, coherence, tone, and factuality across varied domains
• Ensured alignment with user intent and safety protocols

Nov 2022 – Jun 2024 · 1 yr 7 mos

• GenAI RAG-based Customer Support Chatbot
• Built and deployed a LangChain + GPT chatbot for policy, EMI, and payment-related queries.
• Integrated with vector DB for fast document retrieval and multi-turn conversation support.
• Deployed via FastAPI on AKS with Cosmos DB for history storage and monitoring via Prometheus + Grafana.
• Achieved 30% improvement in response accuracy through retrieval and prompt optimization.
• Merchant Transaction Drop Model
• Developed a gross merchant transaction drop model using XGBoost, achieving 74% AUC in the out-of-time sample.
• Credit Risk Scorecards & Loan Underwriting
• Built new credit risk scorecards for top-up and renewal loans, leveraging stacked models to achieve 75% AUC with improved stability.
• Boosted model performance by 5% by integrating third-party data such as location-based and SMS-derived features.
• Designed a rejection strategy that increased loan approvals by 20% while reducing bad rates by 15%.
• Large-Scale Feature Engineering & Data Pipelines
• Optimized the merchant transaction pipeline, processing over 2000 features from more than 20 billion records to enhance efficiency and scalability.
• Engineered over 3000 features from repayment behavior, location data, and app usage, using CSI, IV, and correlation metrics for effective feature selection.
• Model Deployment & Monitoring
• Deployed model pipelines using PySpark on Airflow, reducing execution time by 30%.
• Built a comprehensive SQL dashboard for real-time model monitoring and validation, reducing manual tracking effort by 25%.
• Designed model monitoring pipelines for three models, tracking PSI, AUC, and bad rates, with results stored in AWS S3 using PySpark.
• Key Skills
• Credit Risk Modeling and Scorecards
• XGBoost, Stacked Models, Feature Engineering
• PySpark, Airflow, SQL, AWS S3
• Model Monitoring (PSI, AUC, Bad Rates)
• Loan Default Prediction and Risk Strategy
• Large-Scale Data Processing (20B+ Records)

Mar 2022 – Oct 2022 · 7 mos

• AI & Machine Learning for Credit Risk & Lending
• Fine-tuned a BERT model to detect EMI bounce occurrences, improving loan processing efficiency by 30%.
• Developed a loan amount estimation model with 15% MAPE, boosting prediction accuracy.
• Built credit default prediction models, achieving 76% AUC and increasing productivity by 25%.
• Designed risk models for loan approvals and collections, improving approvals by 10% and reducing defaults by 3%.
• Deployed models using FastAPI and optimized collection strategies for better performance.
• Model Explainability & Deployment
• Developed explainability frameworks using Shapley values and ALE plots to ensure model transparency.
• Deployed models via FastAPI, efficiently handling over 10,000 daily requests.
• Built loan approval models using bureau and banking data, enabling real-time scoring and decile-based segmentation.
• ETL & Feature Engineering
• Built ETL pipelines using Spark, Python, and AWS, generating over 10,000 features and reducing production turnaround time by 30%.
• Engineered more than 3,000 features from bureau, GST, and banking data, automating daily data pipelines.
• Automation & Data-Driven Insights
• Automated credit underwriting processes, reducing review time by 200 minutes across three business verticals.
• Conducted bureau scrub analysis for over 185,000 customers, optimizing sales office locations and loan amounts.
• Designed a SQL-based data model in dbt, reducing data retrieval time from days to hours.
• Key Skills
• Machine Learning (XGBoost, LightGBM, BERT)
• Credit Risk Modeling and Loan Default Prediction
• Model Deployment (FastAPI, Docker, CI/CD)
• Explainability (Shapley, ALE, LIME)
• ETL and Feature Engineering (Spark, AWS, SQL, dbt)
• Real-Time Scoring and Decile Segmentation

Jan 2020 – Feb 2022 · 2 yrs 1 mo

• Credit Risk & Consumer Loan Modeling
• Developed a probability of default model for consumer loans using XGBoost, improving risk prediction accuracy.
• Created a business rule engine for consumer loans to automate decision-making and enhance loan processing efficiency.
• Data Pipeline Engineering
• Built a data pipeline for parsing GST JSON into 10 structured tables, enabling comprehensive financial analysis.
• Engineered a pipeline to parse bureau XML into 4 tables, improving the processing of credit history data.
• Developed a system to extract bank statement data into 3 structured tables, allowing detailed transaction analysis for consumer loan underwriting.
• Data Enrichment & Feature Creation
• Designed an address matching algorithm for KYC validation, improving data verification accuracy.
• Built a photo matching algorithm using a three-layer Convolutional Neural Network (CNN) to compare Aadhaar card photos with live photos, enhancing identity verification.
• Created a wide range of features from bureau data, GST data, ITR data, and bank statements to enrich datasets and improve model training outcomes.
• Expert Scorecards & Model Deployment
• Developed expert scorecards using GST, banking, and financial data to enhance creditworthiness assessment.
• Deployed machine learning models using REST APIs via FastAPI and Flask for seamless integration into production systems.
• Version Control & CI/CD
• Managed version control and continuous integration through GitLab, streamlining the model development and deployment process.
• Key Skills
• Credit Risk Modeling (Probability of Default, Risk Prediction, XGBoost)
• Business Rule Engine Development (Policy Implementation)
• Data Pipeline Engineering (ETL for GST, Bureau, Bank Statements)
• Feature Engineering (Bureau Data, Financial Data)
• Model Deployment (REST API, FastAPI)
• CI/CD and Version Control (GitLab, Deployment Automation)

Jul 2019 – Dec 2019 · 5 mos · Kolkata, West Bengal, India

• Digital Solutions for Construction Industry 🏗️📊
• Designed a digital solution for the construction industry 🏢, enabling daily progress reporting 📅, real-time data visualization 📈, and on-site work cost analysis 💰, reducing manual effort by 20% ⏳.
• Defect Detection with Deep Learning 🤖🏢
• Introduced a deep learning model 💻 using CNN 🧠 for the construction industry to automate defect detection 🛠️ in building inspections 🏚️. Achieved an AUC of 0.78 🎯, accurately identifying and classifying defects 🔍🔧.
• Key Skills 🛠️🚀
• ✅ Digital Transformation & Automation (Process Optimization, Cost Analysis, Reporting) ⚙️💰📊
• ✅ Real-time Data Visualization (Dashboards, Monitoring, Analytics) 📈📊👀
• ✅ Deep Learning for Defect Detection (CNN, Image Classification, Inspection Automation) 🤖🛠️🏚️
• ✅ Computer Vision (Feature Extraction, Object Detection, Quality Control) 🏗️🔍📸
• ✅ AI-driven Construction Analytics (Predictive Maintenance, Anomaly Detection) 🏢📉🔧
• ✅ Data Engineering (ETL Pipelines, Structured Data Processing for Construction) 🔄💾🏗️

Sep 2015 – May 2017 · 1 yr 8 mos · Remote

• 🎯 Provided one-on-one and group training in Python, data analysis, and machine learning for students and professionals.
• 📚 Designed and delivered courses on pandas, NumPy, Matplotlib, Seaborn, and Scikit-learn, focusing on real-world applications.
• 🧠 Mentored learners in data preprocessing, visualization, and statistical analysis, ensuring practical understanding.
• 🛠️ Conducted hands-on workshops on EDA, feature engineering, and ML models, building a strong foundation for aspiring data scientists.