AI Training and Evaluation - Knowledge Base QA System
Built and evaluated a knowledge-based intelligent conversational system using local LLMs and vector databases. Constructed and curated a domain-specific FAQ knowledge base, split and embedded documents, and tested question-answering workflows for accuracy improvements. Performed iterative evaluation and prompt/parameter optimization based on simulated question sets and user interaction evaluation. • FAQ documents were parsed and embedded for effective retrieval. • Retrieval-augmented generation (RAG) techniques facilitated precise response generation. • Simulated 50 diverse question-answer pairs for benchmarking and quality assessment. • Achieved accuracy improvements from 75% to 90% through data labeling and iterative tuning.