About Reflect

Reflect answers one question: given this résumé and this job posting, is it worth applying — and if it is not, what would have to change.

It is free, there is no account, and it is made by one person. This page is about why it works the way it does — what it was built to notice, the choices that followed from that, and where its boundaries are.

Cost
Free. No account, no sign-up, no advertising, and nothing here is sold.
Built for
Software engineering and technical roles.
Last checked
5 September 2026. Every claim on this page was read against the running product on that date.

Why it exists

People read a job posting the way you read a checklist — the years, the degree, the named tools — find their own words on their own résumé, and conclude they are a plausible candidate. Then the application goes nowhere, and nothing in the process tells them which of those things actually counted.

A tool named on a résumé is a claim. It is not evidence that the work was done. That gap is what Reflect was built to look at.

I had tried two other ways of getting at it first. One was to match harder: I built a system that pulled in job listings and scored them against a résumé, then filtered them more and more strictly. The results got fewer without getting more trustworthy.

The other was to hand the question to a language model, which wrote well but did not always answer the same way twice, and was better at making a résumé read closer to a posting than at working out whether the work was there. Neither attempt said the approach was wrong. Both said it could not settle the thing I wanted to know.

What shaped it

Four choices follow from that. Each one buys something and gives something up, and the second half is the part worth knowing before you use the result.

The score is computed, not written.

A language model can be asked the same question twice and answer differently. That is fine in a sentence and not fine in a verdict, so the score, the tier and the decision are arithmetic, and the model only writes the explanation around them — which is why the wording of a report changes between runs and the result does not.

In exchange, Reflect cannot smooth over a result it would rather not give.

Where a skill appears changes what it is worth.

The alternative is to count a word wherever it turns up, which is what most tools within a candidate’s reach still do, and what I had built before. Reflect weighs the same skill differently depending on where the document records it — How It Works shows what that looks like on a real page.

The trade-off is that a résumé which lists real skills without showing them scores lower here than a keyword match would have scored it.

It declines rather than guesses.

When two documents do not support a reading, the choice is between a low-confidence number and no number at all. A low-confidence number is still remembered as a number.

So Reflect will sometimes take two documents and return no score.

There is no account and no price.

Every version of this that collects something — an email to send the report to, an account to keep a history in — is better at growing and worse at telling you the truth, because a product that needs you to come back has a reason to flatter you.

What that costs you is permanence: nothing is saved to a profile, and closing the tab ends access to the analysis.

Where it falls short

Three boundaries are worth knowing before you lean on the result.

It can only read what the document records. Reflect assesses the résumé in front of it, not the career behind it. Work you have genuinely done but did not write down is not evidence it can weigh, so a reading can be accurate about the document and still fall short of the person. That is the boundary of the choice above rather than a fault in it, and it is worth knowing before you read a low score as a verdict on your ability.

An analysis does not persist. It is reachable only from the browser tab that produced it, because there is no account to look it up from. If you want to keep one, download the PDF before you close the tab.

What is submitted is kept for 90 days. It used to be kept with no expiry, and this page said so. Now each analysis is deleted after 90 days, and anyone can ask for theirs to go sooner; Privacy & Trust sets out how. It has been fixed.

Who makes it

Siddhant Thakur

Builds and maintains Reflect

Reflect is made and maintained by one person rather than by a company. The choices on this page were made deliberately rather than inherited from a process, which is why each one can be explained and why there is someone to explain it.

It is still being worked on, and the section above is the current list of what I know is wrong with it.

Every claim here is meant to be checkable against the product itself. If a line on this page does not match what Reflect actually does, it is worth correcting — support@reflectcareer.com.