LLM Applied Engineer2024 - 2025
RAG-powered Interviewer System
PyTorchLangChainFAISSFastAPIPostgreSQL
Summary
Built a structured interview system with RAG for role-specific retrieval, follow-up reasoning, and explainable feedback.
Problem
Interview knowledge changes quickly and responses lacked depth and consistency.
Solution
Designed multi-source retrieval with re-ranking and question decomposition, plus role profiles for scoring rubrics.
Impact
- Improved question coverage and reduced content maintenance cost.
- Delivered consistent evaluations with explainable reports.
Architecture
Retrieval + re-ranking + follow-up planning, combining question bank and company knowledge base.
Key Techniques
Key Techniques
Hybrid retrieval, multi-source re-ranking, question decomposition, and scoring rubric generation.