The dangerous failure in AI-assisted CV writing is often not an obviously false statement. It is a polished sentence that sounds plausible and quietly says more than the source material supports.

I built CV Aligner around the opposite behaviour: if the evidence is missing, expose the gap.

Requirement-by-requirement evidence

The product compares a CV with a real job description requirement by requirement. Each assessment shows the source CV lines that support it, followed by reframes that must remain tied to those lines.

Employers, dates and metrics are treated as immutable. They are not creative-writing inputs.

This produces a narrow contract:

• Surface relevant evidence that is easy to miss.
• Suggest clearer wording for evidence that already exists.
• Preserve factual anchors.
• Leave genuine experience gaps visible.
• Require the applicant to verify every suggestion.

The model is not the whole product

The front end uses React, Vite and TypeScript. Anthropic models are called server-side, but model choice is less important than the surrounding constraints.

For this kind of system, the design questions are:

1. Which fields may the model transform?
2. Which fields must never change?
3. What source evidence must accompany an output?
4. What should the system do when evidence is absent?

A fluent output without those boundaries is not enough.

Honest limitations

CV Aligner cannot manufacture experience that is not in the CV. It cannot decide whether a reframe accurately represents someone's work. That judgement remains with the applicant.

I also do not yet have first-party evidence that it improves interview or callback rates. The current testable claim is narrower: users can see how each suggestion relates to their source CV and where support is missing.

The evidence report is free and needs no account or payment. Optional application packs are one-off purchases rather than subscriptions.

Try it at cv-aligner.com.

Where have explicit model constraints improved trust in a product you built, and where did they simply add friction?