Job Discovery and Eligibility Pipeline
A pipeline that finds jobs I can actually take from Nigeria. It reads each full posting, decides eligibility with a rules-based gate that stores the exact sentence behind every verdict, and only then scores skills fit. Results go to a daily Telegram digest and an admin dashboard.
The Problem
Finding jobs I could actually take from Nigeria was slow, and most of the time went on roles that turned out to be closed to me. A posting can look like a good match and only reveal in its fine print that it offers no visa sponsorship, requires onsite work abroad, is locked to certain locations, or has already passed its deadline. Searching job boards by hand meant reading whole postings just to find that out. A strong skills match is worthless if the role is closed to you. What I needed was a system that checks whether a role is open to me first, and shows me the exact sentence it based that decision on, so I could trust the result without re-reading every posting.
What I Built
I built a pipeline that checks eligibility before it scores anything, so a high skills match can never hide a lockout. It runs daily, searches about 50 queries plus the Greenhouse, Lever, Ashby and Workable APIs and a few job boards, removes duplicates, and reads the full text of the best candidates. The core is a rules-based eligibility gate, not an LLM. It targets two tracks: Remote: the posting explicitly says the role is remote and open to worldwide, Africa or Nigeria. Sponsored relocation: the role is in Canada, the UK, Ireland or Germany, and the posting itself says visa sponsorship is available. Every role gets one of four verdicts: eligible for remote, eligible for sponsored relocation, ineligible, or needs review. The gate splits each posting into sections and ignores privacy and equal-opportunity boilerplate. An explicit “no sponsorship” counts against a role anywhere in the text, but a positive statement only counts in the description, requirements or benefits. Anything uncertain goes to a review queue with a reason code instead of being guessed. I chose rules over an LLM so every decision traces to an exact sentence. That sentence is stored with the verdict, so I can check why the system decided what it did. Only after the gate does the pipeline score skills fit from 1 to 10. Results go into Supabase, a daily Telegram digest, and an admin dashboard where I review roles and mark them Interested, Applied or Passed.
The Result
351 to 55: in a single run, the system searched across APIs, company career-system feeds and job boards, collected 351 raw postings, and narrowed them to 55 worth keeping, with no errors. Every decision has a receipt: each verdict stores the evidence behind it, including the section of the posting it came from, so nothing is a black box. Lockouts caught before wasting time: the gate flagged 25 roles as closed to me, each with stored evidence such as an explicit "no visa sponsorship" statement or an onsite requirement abroad. Precision over volume: on 31 October, of 201 re-reviewed roles, only 2 passed as clearly eligible and the rest went to a review queue. That is deliberate, because a false "eligible" costs an evening of applying. More than 90 automated tests cover the eligibility rules, including regression tests for the false positives I found along the way.