Shasank Kamineni

Built a healthcare copilot using Python, FastAPI, React, LangChain, and AWS Bedrock to help claims analysts retrieve policy evidence and authorization guidance via conversational workflows. Structured FHIR resources and X12 transactions into OpenSearch indexes and Pinecone vector stores, reducing manual policy research effort. Orchestrated multi-agent workflows with LangGraph using Claude and OpenAI GPT models for automated evidence gathering, exception routing, and reviewer collaboration across authorization decisions. Implemented RAG pipelines combining semantic search and contextual retrieval to improve response relevance. Automated document ingestion with PySpark and Airflow and tracked experiments with MLflow, increasing processing throughput. Added governance controls including validation guardrails, audit logging, and role-based access to strengthen PHI compliance.

Engineered an AI-powered support platform using Python, FastAPI, and LlamaIndex to provide contextual troubleshooting recommendations to service representatives. Integrated ServiceNow ticket histories and Salesforce Service Cloud records to shorten investigation cycles by linking recurring incidents to validated resolutions. Deployed scalable recommendation services on Azure AKS with PostgreSQL and coordinated secure interactions between support applications and backend intelligence components. Used Apache Kafka event streams for near real-time synchronization across knowledge repositories and support workflows. Benchmarked semantic retrieval with FAISS to improve knowledge discovery, and implemented evaluation workflows to incorporate analyst feedback and improve AI-assisted resolution adoption.

Built fraud analytics solutions using Python, Flask, XGBoost, and Scikit-learn to identify suspicious transaction behaviors and support proactive financial risk management. Performed feature engineering and predictive modeling on customer transaction datasets, improving anomaly detection precision and reducing unnecessary investigation workload. Constructed supervised learning pipelines backed by MySQL sources to generate consistent risk assessments in high-volume environments. Automated SQL-driven data ingestion and transformation, reducing preparation effort and improving reporting readiness. Containerized services with Docker, monitored production with ELK, and reduced incident resolution time by diagnosing bottlenecks via log analytics and performance monitoring.