zzzlip/langgraph-AI-interview-agent

Implements a hierarchical multi-agent architecture where the main LangGraph orchestrator delegates to specialized subgraphs (resume evaluation, algorithm testing, mock interviews), with fine-grained state isolation using TypedDict/Pydantic and Command-based routing for dynamic task control. Multimodal assessment combines video analysis via OpenCV for body language, audio sentiment via DashScope, and RAG-enhanced question generation using LlamaIndex over a 300+ question knowledge base stored in Chroma DB. Integrates DeepSeek and Qwen LLMs, KeyBERT for semantic matching, and exports professional reports via python-docx and WeasyPrint.