I wrote hireburn and put it under MIT. It's a Python CLI, standard library only, with no network calls, so nothing leaves your machine.
Why the avatar part exists
People put off recording an avatar because they picture a studio. You need a window, a phone and twenty minutes, once. After that the avatar asks every candidate the same questions in the same words, so the first interviews can be compared with each other.
The candidate page says they are talking to an AI avatar before they press Start. The scoring rubric always puts a warning next to the HIRE / MAYBE / NO label: the label is not a measurement.
The part where it does not press Send
hireburn drafts LinkedIn outreach and reminders from a template plus a hook line that a person writes after reading the profile. The queue has random pauses, hourly and daily limits, and working hours. It has a STOP button for the first warning or captcha, and a reply tracker.
A human presses Send. hireburn never does.
We got a LinkedIn restriction when our pace was not human. Bots in this area get their accounts limited, and that's why the default limits are what they are.
Where the funnel leaks
The funnel leaks where nobody looks. No-shows, bookings nobody updated, people who got access and never started, decisions that sat for weeks. hireburn shows the loss at each stage and the days until first real work.
It also puts the AI label next to whether that person is still working on day N. In the case this method grew out of, the AI interview labels did not match who was still working weeks later. So the tool puts the label beside the outcome, and you can check it on your own people.
Try it
The demo runs on a synthetic company, and all its numbers are invented and round on purpose.
pip install hireburn
hireburn demo
Telling the candidate is a feature here. Disclosure, consent, a human alternative and human review are built into the page and the rubric, so the lawful path is the default one. There is also a disclosure checklist for NYC Local Law 144, the Illinois AI Video Interview Act, Illinois HB 3773 and the EU AI Act (hiring is high-risk under Annex III). It's a guide, not legal advice.
This is part of the payrollburn series of cases about AI managers in a live company. The method report is at doi.org/10.5281/zenodo.23927437, and the story of the case is on DEV.to: AI hiring in a live company.
Code: github.com/arsentev-ai/hireburn · Software DOI: 10.5281/zenodo.23979629 · Method report: 10.5281/zenodo.23927437 · PyPI: hireburn
Evgenii Arsentev, MD, PhD — author of contextburn and the payrollburn case series; Founder & CEO, Dad4Kids.
AI assistance: drafted and edited with an AI tool; what the tool does is taken from its README and code.
Top comments (1)
Not pressing Send is the most important product choice in this workflow. Finding leads or preparing a tailored draft can be automated, but the final outreach creates reputational and consent consequences that deserve a human decision. I would expose the evidence behind each recommendation as well, so the reviewer can tell why a person was selected before approving contact.