I’ve been testing ChatGPT (GPT-4) to filter inbound resumes for a customer support role. Our stack is HubSpot Service Hub, so I wanted candidates with relevant SaaS/CRM exposure.
I gave it 50 anonymized resumes and a clear prompt: "Identify candidates with 2+ years in a SaaS help desk role, experience with ticketing systems, and any mention of Freshdesk, Zendesk, or HubSpot." It flagged 18 as matches. My manual review found 14 were actually correct. Four were false positives where ChatGPT overstated generic support experience. It missed 2 good candidates who used different terminology.
Overall, for this batch: 82% precision on the matches it returned. Useful for a first pass, but you still have to check its work. Has anyone else tried this? I’m curious about prompt tweaks to reduce the false positives.