Problem
Job alerts arrive in the same inbox as everything else, and the deadline is usually shorter than the attention they deserve. By the time a promising posting has been read and judged, the applicants who needed no thought are already in. The loss is rarely a job that was winnable. It is the ones that never got an application at all.
The second problem is quieter. An application written in a hurry reads like an application written in a hurry, and the reader can tell. More applications is not the fix. Arriving earlier, and arriving with something tailored, is.
Process
The agent watches the inbox, decides what actually matters, and pushes only the urgent items out of email and onto WhatsApp, where there is a chance of acting on them before the posting closes.
From there it pulls the posting and drafts an application against the actual requirements, rather than against one template stretched over everything.
Python throughout: the scraping and the NLP do the retrieval, and the model handles both the triage and the drafting.
Result
Running since August 2025. The target was deliberately narrow: reach the posting before the deadline, and send something worth reading when it does.