Harshith G

I’m an AI/ML Engineer specializing in LLM systems, RAG architectures, agentic workflows, and scalable ML pipelines. I focus on building production-ready AI systems that combine retrieval, reasoning, and automation to solve real business problems. At Stripe, I’ve built LLM-powered internal assistants, hybrid retrieval pipelines, and multi-agent workflows using LangChain, LangGraph, Pinecone, FAISS, and FastAPI. I design modular AI architectures involving embedding services, vector indexing, hybrid retrieval, re-ranking, context assembly, and GPU-backed inference. My work has improved document relevance, reduced latency, and helped automate high-volume dispute and support workflows. Before that, at Mastercard, I developed fraud detection models, built PySpark + Kafka streaming pipelines, and helped design components of real-time ML decisioning systems. This gave me a strong foundation in data engineering, production ML, feature pipelines, and model monitoring. I enjoy working on: • RAG pipelines (hybrid search, embedding strategies, chunking, re-ranking) • Agentic AI systems (LangGraph multi-agent workflow design) • LLM inference optimization (GPU batching, quantization, caching) • Retrieval quality evaluation and drift monitoring • Designing microservices for AI/ML in production • Fraud analytics, anomaly detection, scoring systems • Distributed data pipelines (Spark, Kafka, Databricks) If you’re working on LLM infrastructure, retrieval systems, agentic automation, or AI-driven decisioning, I’d love to connect and share ideas.