Shruti Rawat

Architected an end-to-end RAG-based Document Intelligence system for regulatory filings and client reports using fine-tuned Llama 3/Mixtral, LangChain orchestration, and Pinecone, reducing analyst research time by 55% while improving factual accuracy. Built and scaled a real-time credit risk scoring pipeline on AWS SageMaker with automated drift detection, increasing AUC by 12%. Engineered multi-step AI agent workflows for compliance document review using LangChain and AutoGen, cutting manual review cycles by 45% and strengthening audit trail coverage. Designed prompt optimization and LLM evaluation frameworks (hallucination detection, response quality scoring, and A/B testing), and deployed an NLP-powered market sentiment module (BERT fine-tuning) integrated into portfolio risk models. Led MLOps/LLMOps infrastructure on Docker and Kubernetes (EKS) with Prometheus/Grafana observability, reducing deployment cycle time by 60% and maintaining 99.9% inference uptime. Established AI governa

Designed and deployed a demand forecasting and route optimization system using LSTM and XGBoost (PyTorch/TensorFlow), integrated with logistics platforms via Apache Airflow to reduce delivery costs by 28% and improve ETA accuracy by 22%. Built a production-ready RAG pipeline with LangChain, Hugging Face Transformers, and FAISS for intelligent search across 100K+ reports, improving knowledge retrieval by 40%. Developed LLM-based text classification and document summarization using fine-tuned transformers and prompt engineering, reducing manual report processing workload by 30%. Engineered scalable ML pipelines on Databricks and Azure ML Studio using PySpark, MLflow, and Docker, with automated retraining triggered by data drift detection. Built an anomaly detection system for real-time transaction monitoring (Isolation Forest + Autoencoder), reducing false positives by 35% for fraud prevention. Established model evaluation and tuning frameworks using Precision/Recall/F1 and AUC-ROC, deli