Sai Jaswanth

Architected scalable RAG and agentic AI workflows using Python, LangChain, LangGraph, and Azure OpenAI to automate document parsing and improve retrieval precision by 94%. Built high-throughput vector search pipelines integrating Pinecone, LlamaIndex, and dense embeddings across Delta Lake, reducing hybrid semantic query latency from 3.2s to 450ms for enterprise workloads. Fine-tuned open-source LLMs using PEFT/LoRA and deployed optimized inference endpoints with vLLM, increasing serving throughput by 35% while lowering compute costs. Implemented automated model evaluation and guardrails using Ragas and TruLens, mitigating hallucinations and enforcing strict output compliance to achieve 96% answer faithfulness. Developed async microservices and feature processing APIs using Python, FastAPI, asyncio, and PySpark, streaming real-time telemetry into MLflow and supporting 10,000+ daily platform queries. Orchestrated distributed training and deployment workflows with Kubernetes and Helm cha

Architected and modernized enterprise backend services using Python, FastAPI, and Flask microservices, integrating NLP data extraction pipelines that boosted processing throughput by 40%. Engineered intelligent vector search and semantic retrieval modules using LangChain, PostgreSQL, and Pinecone, reducing structured and unstructured query latency by 45% across client data platforms. Built predictive ML models and automated feature engineering pipelines with scikit-learn and PyTorch, interfacing with real-time data streams to improve anomaly detection accuracy by 28%. Designed async, messaging- and event-driven pipelines leveraging Apache Kafka, RabbitMQ, and Redis pub/sub to process high-volume inference logs with 99.9% system availability. Implemented automated model monitoring and evaluation frameworks using MLflow and pytest to track drift, evaluate accuracy, and enforce strict performance benchmarks across deployments. Streamlined MLOps workflows and automated CI/CD deployment pip