danvics/pediatric-ai-scribe-v3
1
0
Fork 0
Code Issues Pull requests Projects Releases 14 Packages Wiki Activity Actions

fix(notes): voice → AI note now produces HTML the editor can render

Browse source 
Symptom Daniel reported: "note recording, not working nor going into
textbox". Root cause was on the server side — /api/notes/from-voice
asked the AI for HTML in its prompt, but real-world models return
markdown ~25% of the time. Tiptap's setContent only renders HTML;
markdown comes through as literal text or a partial render, looking
like the textbox didn't fill.

Server (src/routes/notes.js):
  • New toHtmlBody() helper. If the AI returned real HTML (any
    block tag), pass through. Otherwise run through `marked` so
    markdown becomes <p>/<h*>/<strong>/etc.
  • Strips ```json / ```html code fences before JSON parsing.
  • Stricter JSON-recovery: only accepts {title|body} parsed shape;
    falls back to wrapping the AI's full reply via toHtmlBody().
  • Final guard: if body would be empty after sanitisation, wrap
    the raw transcript so the user can at least edit it manually.

Client (public/js/notes.js):
  • applyGeneratedNote prefers _editor.commands.setContent over a
    full remount — avoids the toolbar-reattach flicker + the brief
    window where the body looked empty.
  • Logs to console when the editor target is missing or Tiptap
    setContent throws, so a future regression is greppable.

Plus the two infra fixes Daniel approved earlier in the same
session — keeping them in this commit since they're already
deployed and tested:

  • src/db/database.js: cleanup interval handle exposed; server
    shutdown now clearInterval()s it before pool.end(). Removes
    the SIGTERM → 9-second-hang → Docker SIGKILL race.
  • src/routes/audioBackups.js: switch multer.memoryStorage() to
    diskStorage with cleanup. 10 concurrent 25 MB uploads no
    longer pin 250 MB of RAM. Identical user-visible perf since
    upload is wire-bound.

New dep: marked@latest (used server-side only in toHtmlBody).
This commit is contained in:
Daniel 2026-04-25 01:35:09 +02:00
parent b0e553c02a
commit 41edc46971
9 changed files with 132 additions and 23 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

0
.codex Normal file
View file

17
package-lock.json generated
View file

@ -1,12 +1,12 @@
{
"name": "pediatric-ai-scribe",
"version": "6.52.0",
"version": "6.52.1",
"lockfileVersion": 3,
"requires": true,
"packages": {
"": {
"name": "pediatric-ai-scribe",
"version": "6.52.0",
"version": "6.52.1",
"dependencies": {
"@marp-team/marp-cli": "^4.3.1",
"@marp-team/marp-core": "^4.3.0",
@ -28,6 +28,7 @@
"helmet": "^8.0.0",
"jsonwebtoken": "^9.0.2",
"mammoth": "^1.8.0",
"marked": "^18.0.2",
"multer": "^1.4.5-lts.1",
"node-pg-migrate": "^7.7.0",
"nodemailer": "^8.0.5",
@ -5698,6 +5699,18 @@
"integrity": "sha512-25GUs0yjS2hLl8zAemVndeEzThB1p42yxuDEKbd4JlL3jiz+jsm6e56Ya8B0VREOkNxLYB4TTwaoPJ3ElMmW+w==",
"license": "MIT"
},
"node_modules/marked": {
"version": "18.0.2",
"resolved": "https://registry.npmjs.org/marked/-/marked-18.0.2.tgz",
"integrity": "sha512-NsmlUYBS/Zg57rgDWMYdnre6OTj4e+qq/JS2ot3KrYLSoHLw+sDu0Nm1ZGpRgYAq6c+b1ekaY5NzVchMCQnzcg==",
"license": "MIT",
"bin": {
"marked": "bin/marked.js"
},
"engines": {
"node": ">= 20"
}
},
"node_modules/math-intrinsics": {
"version": "1.1.0",
"resolved": "https://registry.npmjs.org/math-intrinsics/-/math-intrinsics-1.1.0.tgz",

1
package.json
View file

@ -36,6 +36,7 @@
"helmet": "^8.0.0",
"jsonwebtoken": "^9.0.2",
"mammoth": "^1.8.0",
"marked": "^18.0.2",
"multer": "^1.4.5-lts.1",
"node-pg-migrate": "^7.7.0",
"nodemailer": "^8.0.5",

26
public/js/notes.js
View file

@ -616,11 +616,29 @@
}
function applyGeneratedNote(title, body) {
$('note-title').value = title;
var titleEl = $('note-title');
var container = $('note-body-editor');
if (!container) return;
if (_editor) { _editor.destroy(); _editor = null; }
mountTiptap(container, body);
if (!titleEl || !container) {
console.error('[notes] applyGeneratedNote: editor not mounted; title/container missing');
return;
}
titleEl.value = title || 'Voice note';
// Prefer Tiptap's native commands.setContent over a remount when an
// editor already exists — remounting on every voice take left the
// toolbar reattaching, which has caused flicker + a brief window
// where the body looked empty.
if (_editor && typeof _editor.commands === 'object') {
try {
_editor.commands.setContent(body || '<p></p>', { emitUpdate: true });
_dirty = true;
return;
} catch (err) {
console.warn('[notes] Tiptap setContent failed, rebuilding:', err && err.message);
}
}
if (_editor) { try { _editor.destroy(); } catch (e) {} _editor = null; }
mountTiptap(container, body || '<p></p>');
_dirty = true;
}

2
public/sw.js
View file

@ -4,7 +4,7 @@
// API calls always fresh (critical for medical data accuracy)
// ============================================================
var CACHE_NAME = 'pedscribe-v12-notes5';
var CACHE_NAME = 'pedscribe-v12-notes6';
var SHELL_ASSETS = [
'/',
'/index.html',

4
server.js
View file

@ -363,6 +363,10 @@ function shutdown(signal) {
// Close the Postgres pool so pending queries finish/reject cleanly.
try {
var dbMod = require('./src/db/database');
// Clear the hourly cleanup interval first — otherwise its handle
// keeps the event loop alive past process.exit(0) and Docker
// ends up SIGKILL'ing us mid-shutdown at the 10s hard limit.
if (dbMod && dbMod._cleanupInterval) clearInterval(dbMod._cleanupInterval);
if (dbMod && dbMod.pool && typeof dbMod.pool.end === 'function') {
await dbMod.pool.end();
console.log('[shutdown] DB pool drained.');

9
src/db/database.js
View file

@ -464,7 +464,11 @@ async function cleanupExpired() {
if (sess.rowCount > 0) console.log('[DB] Cleaned up ' + sess.rowCount + ' expired sessions');
} catch(e) { /* table may not exist yet */ }
}
setInterval(cleanupExpired, 60 * 60 * 1000); // every hour (audio backups are 24h)
// Interval handle is exposed so server.js can clearInterval() it during
// graceful shutdown — otherwise Node's event loop stays alive past the
// 9-second shutdown deadline waiting for the next tick of this timer,
// and Docker SIGKILLs us mid-cleanup at the 10s hard limit.
var _cleanupInterval = setInterval(cleanupExpired, 60 * 60 * 1000); // every hour (audio backups are 24h)
setTimeout(cleanupExpired, 10000); // also run 10s after startup
// Query helpers
@ -513,4 +517,7 @@ function convertPlaceholders(sql) {
});
}
// Expose the cleanup-interval handle so the shutdown path can clear it.
db._cleanupInterval = _cleanupInterval;
module.exports = db;

37
src/routes/audioBackups.js
View file

@ -5,6 +5,9 @@
var express = require('express');
var router = express.Router();
var fs = require('fs');
var os = require('os');
var path = require('path');
var zlib = require('zlib');
var multer = require('multer');
var db = require('../db/database');
@ -12,23 +15,49 @@ var { authMiddleware } = require('../middleware/auth');
var cryptoUtil = require('../utils/crypto');
var { serverError } = require('../utils/errors');
// 25MB upload limit (same as transcribe)
var upload = multer({ storage: multer.memoryStorage(), limits: { fileSize: 25 * 1024 * 1024 } });
// 25MB upload limit (same as transcribe). DiskStorage so 10 concurrent
// uploads don't pin 250MB of RAM at once — multer streams to a temp
// file as bytes arrive, the route reads from req.file.path. Identical
// user-visible perf (the upload is bottlenecked on the wire either
// way), much flatter memory profile.
var TMP_DIR = path.join(os.tmpdir(), 'pedscribe-audio-uploads');
try { fs.mkdirSync(TMP_DIR, { recursive: true }); } catch (e) {}
var upload = multer({
storage: multer.diskStorage({ destination: TMP_DIR }),
limits: { fileSize: 25 * 1024 * 1024 },
});
router.use(authMiddleware);
// ── POST save audio backup (compressed) ──────────────────────────────────
router.post('/audio-backups', upload.single('audio'), async function(req, res) {
console.log('[AudioBackup] POST received, hasFile=' + !!req.file + ', user=' + (req.user ? req.user.id : 'none'));
// Always remove the temp file when this handler returns — success or
// failure path. The Buffer-based version had no file to clean up;
// diskStorage means we own the lifecycle.
var tmpPath = req.file && req.file.path;
function cleanup() {
if (tmpPath) {
fs.unlink(tmpPath, function(unlinkErr) {
if (unlinkErr && unlinkErr.code !== 'ENOENT') console.warn('[AudioBackup] tmp cleanup failed:', unlinkErr.message);
});
tmpPath = null;
}
}
try {
if (!req.file) return res.status(400).json({ error: 'No audio file' });
var module = req.body.module || 'encounter';
var originalSize = req.file.size;
// Read the temp file once into a Buffer — gzip needs it in one shot
// anyway. For 25MB max this is a single, transient allocation per
// request instead of a persistent 25MB Buffer pinned by multer.
var raw = await fs.promises.readFile(tmpPath);
// Gzip compress, then AES-256-GCM encrypt the audio data.
var compressed = await new Promise(function(resolve, reject) {
zlib.gzip(req.file.buffer, { level: 6 }, function(err, result) {
zlib.gzip(raw, { level: 6 }, function(err, result) {
if (err) reject(err); else resolve(result);
});
});
@ -45,6 +74,8 @@ router.post('/audio-backups', upload.single('audio'), async function(req, res) {
res.json({ success: true, id: result.lastInsertRowid, originalSize: originalSize, compressedSize: compressed.length });
} catch (e) {
return serverError(res, 'AudioBackup save', e, 'Could not save audio backup');
} finally {
cleanup();
}
});

59
src/routes/notes.js
View file

@ -13,6 +13,28 @@ var logger = require('../utils/logger');
var cryptoUtil = require('../utils/crypto');
var { callAI } = require('../utils/ai');
var { wrapUserText, INJECTION_GUARD } = require('../utils/promptSafe');
var { marked } = require('marked');
// Convert AI output to clean HTML for Tiptap. Tiptap's setContent
// only handles HTML — markdown comes through as literal text. Real-
// world models ignore "output HTML only" prompts ~25% of the time
// and return mixed markdown/HTML. We normalise both:
// • Body looks like HTML (has tags) → return as-is.
// • Body looks like markdown OR plain text → run through marked.
function toHtmlBody(s) {
if (!s) return '';
var t = String(s).trim();
// Strip ```html / ``` code fences a model sometimes wraps the body in.
t = t.replace(/^```(?:html|markdown|md)?\s*/i, '').replace(/\s*```$/i, '');
// Already HTML if there's a real tag.
if (/<(p|div|h[1-6]|ul|ol|li|strong|em|b|i|u|s|a|blockquote|code|pre|br|hr|span)[\s>]/i.test(t)) return t;
// Markdown → HTML. marked handles plain text fine — paragraphs become <p>.
try { return marked.parse(t, { breaks: true }); }
catch (e) {
// Last-ditch: wrap the raw text in <p> so the editor at least shows it.
return '<p>' + t.replace(/</g, '&lt;').replace(/>/g, '&gt;').replace(/\n{2,}/g, '</p><p>').replace(/\n/g, '<br>') + '</p>';
}
}
router.use(authMiddleware);
@ -234,28 +256,41 @@ router.post('/notes/from-voice', async function (req, res) {
{ role: 'user', content: userContent }
], opts);
// The model sometimes wraps JSON in markdown fences or adds text
// before the {. Strip anything before the first { to recover.
// Models sometimes wrap JSON in ```json fences, prepend "Here you go:",
// or close the JSON early. Strip leading prose and any trailing text
// after the last } to recover.
var raw = (result.content || '').trim();
raw = raw.replace(/^```(?:json)?\s*/i, '').replace(/\s*```$/i, '');
var jsonStart = raw.indexOf('{');
if (jsonStart > 0) raw = raw.substring(jsonStart);
var jsonEnd = raw.lastIndexOf('}');
if (jsonEnd > -1 && jsonEnd < raw.length - 1) raw = raw.substring(0, jsonEnd + 1);
var parsed;
try { parsed = JSON.parse(raw); }
catch (e) {
// Fall back to raw transcript as body if the model didn't cooperate.
parsed = {
title: transcript.split(/[.\n]/)[0].substring(0, 80) || 'Voice note',
body: '<p>' + transcript.replace(/</g, '&lt;').replace(/>/g, '&gt;').replace(/\n{2,}/g, '</p><p>').replace(/\n/g, '<br>') + '</p>',
};
var parsed = null;
try { parsed = JSON.parse(raw); } catch (e) { /* fallback below */ }
var title, body;
if (parsed && (parsed.title || parsed.body)) {
title = String(parsed.title || '').trim();
body = toHtmlBody(parsed.body || '');
} else {
// Model didn't cooperate at all — treat the whole response as the
// body and derive a title from the first sentence of the transcript.
title = transcript.split(/[.\n]/)[0].trim().substring(0, 80) || 'Voice note';
body = toHtmlBody(result.content || transcript);
}
// Final guards: never send back an empty body — at minimum show the
// raw transcript wrapped in <p> so the user can edit it manually.
if (!body || !body.replace(/<[^>]+>/g, '').trim()) {
body = '<p>' + String(transcript).replace(/</g, '&lt;').replace(/>/g, '&gt;') + '</p>';
}
if (!title) title = 'Voice note';
res.json({
success: true,
title: (parsed.title || '').substring(0, 200),
body: parsed.body || '',
title: title.substring(0, 200),
body: body,
model: result.model,
});
logger.audit(req.user.id, 'note_from_voice', 'AI-generated note from voice', req, { category: 'notes' });