karthikeya3342/careeros

CareerOS is a self-running job search assistant for college students, new graduates, and developers who want to spend less time on applications. It runs as a local server on your computer and uses multiple AI agents working in sequence to handle the repetitive parts of a job hunt: finding new postings, writing tailored resumes, and preparing you for interviews. Every morning at 8 AM the system automatically searches job boards including LinkedIn, Indeed, Glassdoor, ZipRecruiter, and several Indian regional platforms (Naukri, Apna, Shine, Foundit). It compares those listings against your saved profile and emails you a daily digest of the top matches along with a spreadsheet attachment. You can star the jobs you care about so they survive the next day's cleanup. When you pick a job and hit Apply and Optimize, a chain of AI agents starts up. One agent reads the job description, another drafts a LaTeX resume tailored to that specific role using Google's XYZ bullet format (each bullet states what you accomplished, measured by what, and how). A separate verifier agent then runs the draft through an applicant tracking system simulation, scores it, and sends feedback back to the drafter. The loop repeats until the resume scores at least 80 percent, at which point a compiler agent turns the LaTeX code into a PDF. The same run also produces a networking outreach template and a set of interview prep questions specific to that job. Before any of this, CareerOS asks you to paste in your GitHub, LeetCode, or LinkedIn profiles. It crawls those pages to build a factual skill inventory and checks your GitHub commit history to verify which projects you actually wrote, filtering out template forks. The tech stack is a Python FastAPI backend with Playwright for web scraping and a Next.js front end. Setup requires several API keys: Groq for the Llama model, Google Gemini, a Hindsight memory service, and AgentMail for email delivery. The README includes step-by-step instructions for both the backend and frontend.