This project presents a Skill-Routed Multi-Agent Architecture for Intelligent Resume Screening and Adaptive Technical Interviewing. The system combines Large Language Models (LLMs), semantic embeddings, vector memory, and dynamic evaluator orchestration to create an adaptive AI hiring workflow capable of context-aware candidate evaluation and interview generation. Unlike traditional Applicant Tracking Systems (ATS) that rely on static rule engines or keyword matching, the proposed architecture introduces a semantic routing mechanism that dynamically retrieves or synthesizes evaluator agents based on the candidate’s inferred skill topology and job-aligned reasoning requirements. The platform supports:- Resume PDF ingestion with asynchronous background processing- Semantic skill extraction and embedding generation- Candidate vector storage using Qdrant- Dynamic evaluator retrieval and memory reuse- Adaptive evaluator synthesis for novel skill profiles- Candidate-job fit scoring- Explainable evaluation reasoning- Recruiter-facing dashboard workflows The implementation stack includes:- Python + FastAPI- Ollama for local LLM inference- SentenceTransformers for semantic embeddings- Qdrant as vector database- MySQL for structured persistence- React + Bootstrap frontend This repository accompanies the research work on adaptive multi-agent hiring systems and semantic evaluator orchestration.
Das Anirban (Thu,) studied this question.