A career assistant built around evidence
The problem
Résumés are snapshots. They leave things out, compress years of work into a few bullets, and often omit the evidence behind a person's experience.
Matching job-description phrases to résumé phrases can miss relevant experience when the wording differs. AI can compound the problem by guessing where evidence is missing.
An evidence-first approach
Thingamajob starts with ordinary career evidence: historical résumés, project work, public professional pages, career documents and context supplied by the person.
AI can extract candidate claims and surface uncertainty. A plausible statement alone is not enough to establish a career fact.
The Library of Truth
Clean, directly supported facts may be established automatically when the evidence is unambiguous and internally consistent. Routine differences are resolved behind the scenes; unsupported suggestions do not become facts.
When an approved claim is corrected, the active library reflects the replacement while the original decision and provenance remain in history.
Human control and provenance
The person is asked only when genuine judgment or clarification is required. They can correct wording, reject an unsupported suggestion or add context. AI may propose. Evidence supports. The person decides.
Claims retain citations so the person can inspect their supporting sources. Uncertainty and conflicting accounts remain visible instead of being silently rewritten into agreement.
The first proof
For the first formal proof, I used my own career history as Client #1.
The proof builds from ordinary career evidence rather than feeding the system an already-reconciled answer key. After independent extraction and human review, the result can be compared against a separate benchmark for recovered facts, omissions, unsupported claims and unresolved questions.
Job evaluation
The evaluation flow is designed to accept:
- pasted job-description text
- a job-posting URL
- a supported uploaded file
Those inputs are intended to feed one flow: identify the job's requirements, search the reviewed career evidence, explain the evidence state for each requirement and help draft source-backed application material.
The current prototype presents one concise fit paragraph while retaining detailed requirement assessments privately. Required and preferred qualifications keep their distinction. Missing professional qualifications are known gaps only for a confirmed comprehensive career record; incomplete source coverage is not a fabricated negative.
Multi-client architecture
Thingamajob is one reusable product with isolated client workspaces. Each person's evidence, reviewed claims and history belong to that person and must not leak into another workspace.
I am Client #1; a second independent client exercises the same shared product with separate evidence.
Graceful degradation
The product is designed to preserve your work when assistance is temporarily unavailable, explain the interruption in plain language and offer a clear way to continue.
An interruption should pause assistance, not strand the person using it.
Active product
Thingamajob continues to evolve through real job-search workflows. This case study explains the evidence model, human control and product decisions behind it.