Lakshmi Poojitha Vangapalli

Partnered with fraud, AML, and compliance teams to deliver production fraud detection and investigation platforms. Engineered near-real-time scoring of millions of transactions using supervised ML, deep learning, and LLM-based retrieval to support compliant fraud decisioning. Integrated card transactions, ACH/wire transfers, customer profiles, device metadata, and historical fraud labels across AWS data platforms with consistent schema enforcement. Built a hybrid GraphRAG architecture (FAISS, Elasticsearch, Neo4j) for contextual document and graph reasoning, enabling grounded LLM responses. Designed secure multi-agent orchestration with LangGraph, MCP, and tool interfaces to securely interact with enterprise APIs, databases, and knowledge bases. Deployed GPU-enabled inference services on EKS with Triton, implemented CI/CD and governance, and established monitoring dashboards for performance, latency, drift, and data quality. Exposed services via FastAPI/Node.js microservices for real-t

Developed predictive models to identify patient deterioration and hospital readmission risks using structured EHR data. Built HIPAA-compliant AI solutions on Azure ML and AKS with automated training, monitoring, and security controls to support clinical decision-making workflows. Pulled vitals, labs, diagnoses, medications, and claims into a unified analytics layer; engineered features; built models with XGBoost and LightGBM; calibrated and tested stability. Built PyTorch-based sequence models for time-series signals; developed NLP pipelines to extract symptoms and clinical indicators; applied transformer-based models to summarize clinical narratives and surface supporting evidence. Designed retrieval pipelines using Azure AI Search, vector embeddings, and similarity search over approved clinical guidelines to improve contextual recommendations. Deployed low-latency inference endpoints on Azure ML/Triton, containerized workloads on AKS, and implemented RBAC and Key Vault for HIPAA-alig

Built ML-driven analytics to help financial advisors identify client churn risk and analyze portfolio behavior. Developed predictive models (XGBoost, LightGBM, Random Forest) for churn and portfolio risk scoring, including calibration and stability testing. Built time-series forecasting models (LSTM, ARIMA) for asset trend analysis and advisor reporting. Created semantic search and document retrieval systems using sentence-transformer embeddings and FAISS for research reports and advisor notes. Implemented conversational interfaces with Dialogflow CX and integrated ML predictions into advisor-facing dashboards. Leveraged Vertex AI Pipelines for automated model training, validation, deployment, monitoring, and retraining; containerized inference services on GKE; established monitoring for drift and performance; incorporated SHAP explainability for audit and review.