# ResumeRadar

An AI workflow for résumé screening, public-project research, and job-fit assessment

Source: https://works.im0o.top/en/projects/resume-radar/
Date: 2026-09-18
Tags: AI, Full Stack, React, Data Visualization


Reviewing résumés and looking up the projects they mentioned felt repetitive and tedious. I built ResumeRadar in my spare time to speed up that process: have AI read the résumé, research publicly available projects, and put together an initial screening report.

## From résumé to screening report

Applications and résumés come from a Feishu multidimensional table or can be added manually. The workflow reads the résumé, using OCR for scanned documents. It then assesses it against the job description and extracts projects, links, and information that can be checked publicly. Later stages search, verify the findings, and assemble a report.

Résumé extraction, assessment, verification, open-source review, and interview questions are separate stages with saved results. Individual claims can be checked in parallel, and later stages reuse earlier results. I can rerun the stages that need changing.

## Researching the projects in a résumé

When a résumé provides public links or project leads, AI searches for them, opens the pages, and compares what it finds with the résumé. The report keeps sources, excerpts, and verification results so I can follow up.

GitHub review prioritizes repositories explicitly mentioned in the résumé. For a candidate’s own project, it reads the README, recent commits, and source files relevant to the role. For contributions to someone else’s project, it examines the candidate’s pull requests and changes. Account ownership and the candidate’s actual contribution are part of that review.

![Follow public-project leads from the résumé and retain the sources in the screening report](https://image.im0o.top/2026/20261117104440427.webp "Follow public-project leads from the résumé and retain the sources in the screening report")

## Using Jev to help decide who to interview

I set assessment criteria based on the job requirements and résumé quality. The report includes scores, reasoning, and the résumé excerpts behind them, so I can compare candidates and sort the list.

I integrated the evaluation model Jev: the language model writes its analysis and supporting excerpts first, then Jev scores them against the criteria. In my comparisons with the same samples, direct LLM scoring varied between runs, while Jev produced almost identical scores. I’m happy with that consistency.

![In my repeated comparisons of the same samples, Jev scores were almost identical](https://image.im0o.top/2026/20261117104434685.webp "In my repeated comparisons of the same samples, Jev scores were almost identical")

## Interview preparation and further assessment

The report includes the assessment, public-project findings, strengths, and points to follow up on. I use these materials and the job requirements to generate interview questions with AI. Each question has a focus, expected answer points, and follow-ups. I can set question preferences and draw from a question bank I maintain.

I can continue discussing the candidate with the built-in agent, using the résumé and screening report as context. After an interview, I can import meeting notes as material for further assessment.

The report and original résumé can be exported together as a PDF to share with HR or other interviewers.

As the candidate list grew, I built table and board views with saved configurations. I can filter and sort the list, then open a side panel to read a résumé and its report.

Human corrections are saved separately and retained across reruns. I can rerun the stages that need another look and review the history of each run.

