The tricky part in this one is not matching a mention, it is deciding where the handle ends.
Twitter stops a mention at the first character outside <code>[A-Za-z0-9_]</code> and caps the handle at 15 chars. The moment you reuse that regex for another platform the spans go wrong, because the allowed set differs. TikTok handles may contain a dot (<code>@some.user</code>) but may not end on one, so a naive <code>/@[\w.]+/</code> cheerfully swallows the period that ended the sentence.
What has worked for me is: match greedily, then trim from the right while the last character belongs to a "cannot terminate a handle" set.
<pre class="shiki github-dark shaku"><code>function scanHandle(text, i, allowed, cannotEnd) { let j = i + 1 while (j &lt; text.length &amp;&amp; allowed.has(text[j])) j++ while (j &gt; i + 1 &amp;&amp; cannotEnd.has(text[j - 1])) j-- return j &gt; i + 1 ? [i, j] : null}</code></pre>
The scan stays single pass and every platform-specific rule collapses into two Sets, which is much easier to test than one regex per platform.
This is not only an interview thing. <a href="https://tikviewer.org/">TikViewer</a> has to accept <code>@some.user</code>, <code>some.user</code> and a full profile URL as the same input, and the trailing-dot case was the very first bug users reported.
For the overlap half of the problem, sorting matched spans by <code>startIndex</code> and dropping any span that starts inside the previous one is enough. No interval tree needed at this input size.

Where a handle ends is the real edge case, not the regex

On Twitter(now X), we always see mentions, hashtags and links in blue links, like this one below. > Shaku syntax is now supported on https://t.co/Upkm…