Varsha Hemakumar

AI Engineer at Finta

Architected and deployed Generative AI agent workflows using LangChain, GPT-4, and tool-calling frameworks to deliver contextual financial intelligence across $1B+ investor portfolios, boosting CRM engagement by 38% through personalized, multi-turn insights. Implemented and productionized BERT-based intent classification and sentiment analysis models processing 80K+ daily interactions, improving lead prioritization by 42% and accelerating revenue pipeline conversion. Built enterprise-scale semantic search and vector embedding pipelines with OpenAI embeddings, FAISS/Pinecone, indexing 5M+ CRM/LinkedIn profiles, reducing information retrieval latency by 65%. Established compliance-ready AI governance and model risk frameworks ensuring auditability across production models. Developed a Retrieval-Augmented Generation (RAG) platform enabling analysts to query 10M+ records with sub-second responses, cutting manual research by 70% and driving $2.5M+ projected annual uplift.

AI & Data Science Engineer at Goldman Sachs

Engineered enterprise-grade NLP and Generative AI pipelines (GPT-4, BERT, Hugging Face Transformers), processing 50K+ financial/regulatory documents daily, accelerating risk intelligence workflows and improving signal accuracy. Developed credit-risk and fraud detection models using XGBoost and TensorFlow, monitoring 12M+ transactions weekly and reducing false positives by 31%, contributing to ~$6.2M in annual fraud-loss avoidance. Architected a scalable RAG platform indexing 8TB+ of regulatory filings, cutting analyst research time by 70% and improving audit response SLAs by 55%. Automated end-to-end MLOps for 25+ production models (MLflow, Airflow, Docker, CI/CD) increasing release velocity by 40% and ensuring governance. Deployed a real-time GenAI decision intelligence system delivering 35K+ daily predictions on AWS SageMaker/Kubernetes with sub-second latency, supporting $100M+ risk-exposed portfolios.

Machine Learning Engineer at KPIT Technologies

Designed and deployed AI-driven credit risk and fraud detection models using XGBoost, LSTM, and ensembles, analyzing 10M+ loan/transaction records to improve approval precision by 28% and reduce default exposure by $4M annually; identified 1,200+ fraud patterns monthly in real-time production systems. Built advanced time-series forecasting for treasury/liquidity analytics, optimizing $50M+ in working capital planning and boosting forecast accuracy by 35%. Developed enterprise NLP/Generative AI solutions (GPT-based summarization, sentiment analysis, RAG) processing 600K+ documents annually, cutting manual analysis by 70% and shortening executive reports from hours to minutes. Implemented CV pipelines with OpenCV to process 250K+ KYC/compliance documents, reducing manual verification by 75% and improving identity validation by 40%. Led end-to-end MLOps (MLflow, Docker, Kubernetes, CI/CD) for 30+ models and led the development of an LLM-powered decision intelligence system delivering real