Anwin Mathew Josy

• Optimized MySQL database schemas and designed scalable RESTful APIs, improving performance and reliability under high-frequency request loads.
• Refactored backend architecture and query handling, reducing response time by 20% across critical API endpoints and enhancing overall system efficiency.

The Beru Refinery Assistant is an AI-powered system designed to enable secure and efficient retrieval of technical information from over 1, 000 pages of refinery manuals, supporting accurate query resolution for domain-specific use cases. The system leverages advanced Generative AI techniques, including Large Language Models and Retrieval-Augmented Generation (RAG), to provide context-aware responses.
To ensure optimal performance on resource-constrained environments, the LLaMA 3.1 model was optimized using 4-bit quantization, achieving sub-3-second response times on CPU-based local systems. A hybrid retrieval mechanism combining BM25 and FAISS was implemented to significantly enhance retrieval accuracy, resulting in a 35% improvement in domain-specific query performance. The project also contributed to an innovative system design that led to the filing of a Registered Copyright.