AI Engineer at Red Sky Health

Architected and developed an Agentic RAG system for healthcare claim-denial resolution. Built a hybrid retrieval pipeline using Azure Cosmos DB vector search, BM25, and full-text search to index 249K policies and 4M policy chunks. Implemented agentic prompting/RAG tool workflows to generate search queries, assess evidence, and trigger additional retrieval when needed. Optimized multi-stage retrieval and reranking to improve relevant policy retrieval by 28% and reduce irrelevant evidence sent to downstream LLMs. Integrated claim details, denial codes, procedure/diagnosis codes, and policy evidence into the LLM workflow, reducing manual policy-review effort by 35% and improving consistency of claim resubmission recommendations.

AI Engineer at LexisNexis

Architected a Python multi-agent compliance platform using Google ADK and LangGraph orchestration to process 10K+ documents/month across three enterprise client workflows, reducing manual legal document review time by 45%. Designed and deployed a production agentic RAG backend with contextual embeddings, Pinecone semantic search, and session-aware vector retrieval over 200K+ unstructured legal filings, reducing hallucination rate by 38% and improving retrieval precision by 34% versus keyword-only search. Engineered async consumers on AWS SQS/SNS/S3 with batching, retry logic, and idempotency controls to sustain 99.7% task reliability under peak loads of 5K+ concurrent agent requests. Owned end-to-end MLOps lifecycle by deploying agentic microservices on Kubernetes with Jenkins and Argo CD GitOps, reducing release cycles from weekly to daily with zero rollback incidents over four months. Delivered a stateful Issue Analysis Agent that auto-generates compliance remediation actions, reduci

Software Engineer at Cardinal Health

Integrated Hugging Face LLaMA-2 into a clinical NLP pipeline for SOAP note generation, reducing documentation time by 35% per consultation and lowering note error rate by 22%. Built a GenAI RAG pipeline (LangChain, FAISS, AWS Bedrock) as a FastAPI microservice for ICD-10 retrieval and FDA alert queries, achieving sub-300ms P95 latency across 50+ regional centers. Engineered HIPAA-compliant PySpark ETL pipelines in Databricks orchestrated with Airflow into Snowflake, increasing ingestion throughput by 40% and eliminating three weekly manual refresh jobs. Trained an XGBoost risk stratification model on AWS SageMaker with 0.91 ROC-AUC and deployed it as a REST endpoint to improve high-risk patient identification by 17% over a rule-based baseline. Containerized ML inference services with Docker and AWS ECS, reducing deployment time from 2 days to under 90 minutes using GitHub Actions CI/CD with automated rollback on health-check failure.

Software Engineer at Capgemini

Owned deployment of production ML scoring APIs in Python using scikit-learn and FastAPI, containerized with Docker and orchestrated via Azure ML Pipelines, reducing model serving latency by 28% versus batch scoring. Developed a Random Forest fraud detection pipeline using transactional velocity, geolocation, and device fingerprinting features, reducing false negatives by 35% and preventing an estimated $2.1M in fraudulent FX transactions annually. Engineered PySpark and Databricks SQL preprocessing pipelines for 50M+ daily transactions, cutting feature engineering runtime from 6 hours to under 90 minutes and serving as technical lead for a three-person data engineering pod. Applied SVM on high-dimensional KYC behavioral features to improve anomalous profile detection by 33% and reduce false-positive escalations by 19%. Automated 50+ Excel reporting workflows into Python/SQL pipelines, reducing turnaround from 3 days to under 6 hours and surfacing $1.2M in previously undetected operatio