Arthi Kotha

Designed and deployed production-grade agentic AI workflows using LangGraph state machines and execution graphs to coordinate planning, reasoning, memory, tool invocation, and multi-step execution. Built agent-tool pipelines integrating LLMs with enterprise REST APIs, databases, and Kafka streams using structured function calling and schema-driven contracts. Implemented short-term context and retrieval-based memory patterns using vector stores, hybrid retrieval, reranking, metadata filtering, grounding, and citation-oriented retrieval. Developed FastAPI REST services for scalable LLM/agent inference, added deterministic guardrails, response validation, and human-in-the-loop approvals for safer autonomous decisions. Established governed MCP-based agent integration patterns with RBAC, schema validation, secure API access, and fail-safes. Implemented evaluation and observability for agentic workloads using RAGAS, DeepEval, LangSmith, TruLens, OpenTelemetry, Prometheus, Grafana, and CloudW

Built ETL and data-processing workflows using Python and PySpark to prepare high-quality datasets for predictive modeling and analytics. Developed LLM integration components for document analysis, summarization, and question answering, and delivered FastAPI-based inference services and REST APIs to integrate ML/LLM capabilities into enterprise applications. Implemented vector-based retrieval and metadata filtering strategies to improve contextual relevance, reduce hallucinations, and provide grounded responses. Designed RAG architectures combining embeddings, semantic search, retrievers, and prompt engineering, and implemented event-driven near-real-time workflows using Kafka and Spark Streaming. Built end-to-end ML solutions for training, retraining, and batch prediction, including classification and regression using XGBoost/LightGBM, plus deep learning models in PyTorch (CNN, LSTM, transfer learning). Applied MLOps practices with MLflow and automation with Airflow/Kubeflow, supported

Integrated and optimized Snowflake data models and SQL transformations for AI/ML and RAG workloads, and built PyTorch inference services integrated with backend APIs and data pipelines. Built and fine-tuned PyTorch NLP and predictive models for claims and underwriting use cases using domain-specific insurance vocabularies. Designed secure AWS architectures using VPC, encryption standards, and IAM policies for compliance and data protection. Led ingestion pipelines for structured and unstructured data including contracts, claims files, underwriting documents, and policy PDFs, and developed Databricks notebooks in PySpark/Spark SQL for cleansing, deduplication, enrichment, and aggregation. Implemented end-to-end RAG pipelines (retrieval, context injection, inference, post-processing), built microservices/APIs for internal AI experiences, and established unit/integration/regression testing across retrieval quality and reliability. Reduced claims triage time by 35% using multi-agent summar