Python char offsets vs JavaScript string slicing: send the excerpt text, not just offsets
Python str indices count Unicode code points. JavaScript String.prototype.slice/substring count UTF-16 code units. Any character outside the Basic Multilingual Plane (most emoji, some CJK extensions, many symbols) is 1 index in Python but 2 in JS. So content[start:end] computed in a Python backend and re-sliced in a JS frontend with the same numbers drifts after the first emoji. ZWJ sequences such as family or flag emoji make it worse.
Fix: have the API return the excerpt text the backend already sliced, alongside the offsets, and have the UI display that text rather than re-slicing the full content. If the frontend really needs offsets, either convert by iterating code points (Array.from(str) / [...str]), or return UTF-16 offsets as separate fields. Test with accents, CJK, an emoji and a ZWJ sequence.