During my internship as an AI/OCR Search Engineer, I worked on building production-grade LLM pipelines for structured document extraction and retrieval. My primary focus was designing deterministic extraction systems that generate schema-validated JSON outputs with high reliability and low latency. I developed preprocessing pipelines to reduce OCR noise using image enhancement and rule-based cleaning techniques, which significantly improved extraction accuracy. I implemented Retrieval-Augmented Generation (RAG) systems using LangChain and FAISS to enable semantic document search and grounded responses. In addition, I built FastAPI-based services exposing inference and retrieval endpoints, added input validation and automated tests, and implemented structured output validation using Pydantic to reduce hallucinations. I also created evaluation scripts and dashboards to measure precision, recall, F1 score, latency, and failure patterns, enabling data-driven optimization of the system. Overall, my role involved taking AI models from experimentation to production-ready systems with measurable performance metrics and reliability standards.