An NLP matching engine for temporary work, in production
Co-founded and ran the company that shipped it
Context
Temporary staffing runs on phone calls. A shift opens, someone works through a contact list, and the match is made by whoever answers first rather than by who fits. We built a conversational system that took the request in natural language and matched it against availability and skills.
I co-founded the company and ran it as CEO. That meant the model was mine, and so was the payroll — the two constraints turn out to be related.
Constraint
Two sides, both allergic to the interface. Job seekers arrive worn down by recruiters and job platforms, and expect the next one to waste their time too. On the other side is someone looking for help with a household job, with no technical proficiency and no patience for a form.
Anything that looked like a portal would have lost both of them.
What I built
In 2020, before LLMs, I designed a chatbot that turns a request into a matched candidate nearby in under sixty seconds: no friction, the graphical interface cut to a minimum, plain and friendly language. The onboarding and the job request ran as a conversation flow in Landbot.
In 2021 we shipped iOS and Android apps carrying the same flows, plus a dashboard where the two sides could coordinate the job and settle payment.
What happened
The apps ran for two years with a steady inflow of job seekers, and never became financially sustainable.
In 2025 I rebuilt the platform as a web app to cut running costs, and added automated assessment of the completed job — photo uploads scored by an LLM (Gemini). It is still online, at Nimbo.ai.