Quorum

I built Quorum ("Chairlift"), an agentic resume-screening system where a resume and job description go in, and six specialized AI agents evaluate the fit together, parsing the resume, parsing the role, matching skills, weighing experience, making a decision, and explaining the reasoning in plain terms. It runs on a FastAPI backend, works deterministically by default, and can use Gemini for deeper semantic extraction. It solves a real gap in resume screening: most tools are either rigid keyword filters that reject good candidates, or a single LLM giving a confident score with no visible reasoning. Quorum makes every score traceable to the agent that produced it, and more importantly it knows when not to answer. If a job posting is too vague or a resume can't be parsed, it abstains instead of guessing. What I learned: breaking one judgment into specialized agents forces real clarity about what each signal means, rather than letting everything blur together in one prompt. And designing for abstention was harder than designing for an answer teaching a system to recognize weak inputs and say "I don't know" is a much rarer, harder problem than making it sound confident.