Oct 31, 2026 · @Bogdan Gusiev
The detected_faces table can show both failure types (missed faces and wrong blurs) for InsightFace, RetinaFace and Google Vision. Two gaps remain: a Google Vision scan that finds no faces leaves no row, and EgoBlur never wrote to detected_faces at all.
The app's code is in the private GitHub repository bogdan/realphotos-marketplace. GitHub user @bradleykam already has access.
How the exported transcript of Review #152 (Auto-corrected) becomes Review #153 (Bogdan Gusiev)'s, with PHI masked.
Legend: removed · added · Edited: 24 · Unchanged: 2 · Added: 2
| Turn | Speaker | Change | Text |
|---|---|---|---|
| #01 | Doctor | Edited | こんにちは~~、 |
| #02 | Patient | Edited | はい、 |
Program: OSS VRP URL: https://github.com/googleapis/google-api-ruby-client Vulnerability type: Sensitive data exposure
Component: google-apis-core gem (part of googleapis/google-api-ruby-client)
File: google-apis-core/lib/google/apis/core/http_command.rb, HttpCommand#execute_once
Old (dataset 5) vs New (dataset 6) — averages only.
Links open the admin stats page filtered to that dataset, on production.
| Metric | Old | New | Better for reviewer | Change |
|---|
This document collects real examples from already-reviewed conversations that could help the automatic correction feature catch more mistakes, notes where we don't have enough good examples yet, and lists a few practical things worth keeping in mind for anyone preparing or reviewing transcripts.
After a conversation is transcribed, an AI reads through the whole thing
This explains, in plain terms, what it would take to make the automatic transcript
correction feature (app/utils/transcript_correction.rb) catch more errors than it
does today. There are two very different ways to do this, and they cost very
different amounts of time and money.
After a doctor-patient video call is transcribed, an AI model reads through the
| Today there's one autosave path: any change to reviews, utterances, or the trim marks hasUnsavedChanges dirty, a $effect waits 1s of inactivity, then saveDraft() fires one PUT review_draft with the entire current reviews array + entire utterances array + trim. The response is reconciled back against snapshots (savedReviewSnapshot/savedUtterancesSnapshot) taken at send-time to avoid clobbering clicks made mid-request (MeetingReview.svelte:1003-1052). | |
| Two things make this fragile: | |
| - Snapshot drift: reviewSnapshot() is a string built from decision:start_at:end_at. Anything that changes server-side but isn't reflected identically in that string (rounding, reordering, a field the reconciliation didn't anticipate) makes reviewDecisionKey never re-equal savedReviewSnapshot, so hasUnsavedChanges stays permanently true → the 1s effect refires forever → indefinite "Saving…". | |
| - Backend semantics assume a full snapshot. apply_phi_reviews/apply_bystander_reviews treat absence from the array as an explicit signal ("this m |
| [ | |
| { | |
| "meeting_id": 862, | |
| "original_transcript": [ | |
| { | |
| "end": 9.56, | |
| "text": "こんにちは。医師の島原です。え、本、ご本人様確認のためフルネームと生年月日をお願いします。", | |
| "start": 0.0, | |
| "words": [ | |
| { |