AI product / engineering case study
JobFndr
A useful shortlist starts before the AI score.
I built a remote-job research product that connects candidate profiles with eligible roles, explains the fit, and prepares applicants with source-backed employer research.
Explore JobFndr- My role
- Full Stack Engineer
- Stage
- Live product · soft launch
- Year
- 2026
The problem
A remote label does not mean a role is open to every country. Candidates still have to find current listings, check restrictions, judge fit, and research each employer. A high AI score is unhelpful if the job is stale, duplicated, or unavailable to the person reading it.
My contribution
I built the profile-to-shortlist experience and the systems behind it: source ingestion, eligibility and relevance rules, AI ranking, saved-job workflows, cited company research, and the scan-credit lifecycle. My focus is making the product explain what it found and remain honest when the evidence is incomplete.
The decisions
Check eligibility before asking AI to rank
A language model can find a strong skills match while overlooking an explicit hiring restriction.
Deterministic checks screen location, work eligibility, freshness, duplicates, and role relevance first. AI ranks the surviving candidates; it cannot override a recognized eligibility conflict. This can produce a smaller shortlist, but preserves the meaning of a match.
Build a shared inventory with traceable sources
Fetching everything during each user search makes latency and source quality harder to control.
Approved feeds and employer boards feed a normalized inventory. Each posting retains source and freshness evidence, while bounded fallback discovery fills qualified gaps. The tradeoff is maintaining ingestion and source policies in exchange for a more controlled search path.
Make research claims earn their place
Plausible company prose is easy to generate. Verifiable preparation is harder.
Employer research is tied to actual web citations. Missing evidence produces a retryable or limited result rather than invented company facts. The scan-credit flow also releases a reservation when a run fails or saves no qualifying new match.
Implemented workflow
- 01
Understand the candidate
Review a saved profile or extract résumé fields for the candidate to confirm.
- 02
Retrieve and screen
Find current roles, resolve duplicates, and check the employer’s location and eligibility requirements.
- 03
Explain the shortlist
Show match reasons and skill gaps, then save qualifying opportunities with their original source.
- 04
Prepare and hand back
Research a selected employer with citations. The candidate decides whether to apply on the original listing.
What you can inspect
A live product connecting profile setup, job discovery, explained matches, saved roles, and employer research in one workflow. The engineering result is an inspectable search process with explicit eligibility boundaries and failure states.
Current stage
Live in soft launch. The evidence here covers product behavior and engineering decisions; usage and hiring-outcome metrics are not yet published.