iam-epiphany/resume-agent
A resume chatbot that answers interviewer questions in first person but refuses to make things up without evidence
ResumeMind turns a person's resume, project write-ups, and certificates into a searchable knowledge base so interviewers can ask free-form questions and get first-person answers. Hard facts like numbers and dates must come from retrieved documents, and when evidence is thin, the system is designed to flag its answer as a guess rather than fabricate. It ships with 480 tests and reports latency improvements on a 30-question evaluation set.
What it does
- Resume, certificates, and project descriptions are reorganized into topic-based Markdown documents before being chunked and stored in a searchable knowledge base
- Questions are routed into types like self-introduction, project deep-dive, general technical, HR/behavioral, and resume detail, each with different evidence requirements
- Dense (meaning-based) retrieval and keyword retrieval are combined via RRF fusion; a reranker model only kicks in for ambiguous cases to save time
- After generating an answer, the system checks whether hard facts like numbers, dates, and proper nouns actually appear in the retrieved evidence and belong to the correct project
- On a 30-question interview evaluation set, median end-to-end latency dropped from 13.1s to 4.08s, and 91.6% of expected facts were correctly included in answers
Why it matters
Resume-based chatbots are prone to inventing experience or numbers that don't exist, and this project shows a structural way to reduce that risk through grounding checks and a fact ledger. It's also a concrete reference for how to design retrieval, reranking, and time-budget management in a RAG pipeline.
Terms in this repo
- RAG · An approach where relevant documents are retrieved before generating an answer, using them as grounding evidence
- Dense Retrieval · A search method that finds documents with similar meaning even without exact keyword matches
- RRF fusion · A method for merging ranked results from different retrieval methods into one ranking
- Reranker · A model that re-scores an initial set of retrieved candidates for more precise ordering
- Grounding check · A verification step confirming that facts stated in a generated answer actually exist in the retrieved documents
Repository description (English)
面向"面试官 × 简历主人公"场景的个人简历 RAG 问答 Agent：把简历、证书、荣誉、项目介绍放进知识库，面试官可对简历任意提问（自我介绍、项目深挖、技术八股、HR 素质、简历细节），系统以第一人称自然作答。核心设计是"宽松推理 + 诚实标注"——硬事实（数字/日期/证书名）必须来自检索，证据不足时强制标注"根据现有知识库推测"而非编造。支持多轮追问记忆、检索兜底链、置信度分级回答、访问码闸门与 IP 限流防刷。
Open on GitHubTrending repos
- vorssaint/vorssaint-utilsOne free menu bar app replaces a dozen paid Mac utilities
- Alishahryar1/free-claude-codeA local proxy that lets coding AI agents run on 49 free or cheap model providers instead of one paid service
- freestylefly/awesome-gpt-image-2A library of 532 reverse-engineered prompts that turn GPT-Image2 into a predictable image-making tool
- block/buzzAn open-source workspace where humans and AI agents chat, code, and review in the same rooms
- NousResearch/hermes-agentNous Research's Hermes is an AI agent that gets smarter the more you use it
- virgiliojr94/book-to-skillA tool that turns technical book PDFs into on-demand reference skills for AI coding agents
- VoltAgent/awesome-agent-skillsA single hub collecting over 1000 'how-to' manuals that make AI coding assistants act like experts
- anthropics/claude-plugins-communityA shared shelf where anyone's Claude add-ons get listed for install