Poojitha Singh

Professional with 3+ years of experience developing Machine Learning and Generative AI solutions across fraud detection, NLP, and document intelligence. Skilled in model development, feature engineering, model evaluation, MLOps, RAG, and LLMs. Proficient in Python, SQL, Scikit-learn, XGBoost, Hugging Face, FAISS, LangChain, MLflow, AWS, Docker, and Kubernetes

GenAI Engineer - IBM - TX

(2026-01)

• Contributed to enterprise document ingestion and preprocessing using Python and LangChain, processing 10K+ internal documents through parsing, cleaning, chunking, and metadata enrichment for RAG workflows
• Implemented semantic retrieval components using Hugging Face, Sentence Transformers and FAISS, generating embeddings and maintaining indexes for 50K+ document chunks, improving context relevance for enterprise queries
• Worked with the GenAI team to integrate IBM watsonx.ai and IBM Granite foundation models into RAG workflows, developing prompt templates and grounding logic that helped reduce unsupported responses by approximately 15%
• Developed and enhanced REST APIs using FastAPI, supporting retrieval and LLM inference services and helping reduce average inference latency by approximately 10%
• Supported RAG evaluation and testing using Ragas and Pandas, analyzing retrieval relevance, faithfulness, and answer quality across 500+ test queries to identify prompt and retrieval issues
• Assisted with containerization and deployment using Docker and Kubernetes, packaging GenAI components and supporting development and testing deployments across multiple environments

ML Engineer - Deloitte - India

(2022-07 - 2024-12)

• Developed Python workflows to clean and preprocess large-scale payment transaction data, handling missing values, categorical variables, and data-quality issues to prepare reliable datasets for fraud modeling
• Engineered transaction, customer, device, and behavioral features using Pandas and NumPy, creating model-ready inputs that improved the quality and consistency of fraud-risk predictions
• Built and evaluated fraud classification models using Scikit-learn and XGBoost, tuning hyperparameters and evaluating precision, recall, F1-score, and ROC-AUC; improved fraud detection recall by ~8% through iterative feature and model tuning
• Processed high-volume transaction streams with Apache Kafka and PySpark, supporting near-real-time feature processing and risk-scoring workflows across millions of transactions
• Developed model inference services using FastAPI and REST APIs, enabling downstream applications to submit transaction data and receive fraud-risk predictions with low-latency responses
• Containerized ML services using Docker and supported deployment on AWS, standardizing application environments and reducing deployment-related issues by ~12% across development and testing
• Managed model experiments, versions, parameters, and evaluation metrics using MLflow, improving reproducibility and making it easier for the team to compare model iterations
• Monitored ML pipelines and model behavior using Amazon CloudWatch and SQL, investigating data-quality issues, prediction trends, and production anomalies while collaborating with data engineers and software developers

MS - Electrical and Computer Engineering - University of Massachusetts Amherst

BE - Electrical and Computer Engineering - MS Ramaiah Institute of Technology