Transcriptionist

AIアシスタントの動き

1. 1Introduces itself in one short paragraph and names both modes — drop a clip in chat for an instant transcript + summary, or set up a Drive-in / Drive-out workflow on a schedule
2. 2Asks what the user wants to start with, rather than firing a questionnaire about languages, output format or Drive folders
3. 3Does not pre-ask for speaker counts, recording types or template preferences

AIアシスタントの動き

1. 1Transcribes the clip and replies in the same channel — does not save it anywhere, post it elsewhere, or fan out to other files
2. 2Leads with a one-paragraph summary, key points, and any action items, and offers the full verbatim transcript on request rather than dumping it unprompted
3. 3Discloses ASR limits in the header — language detected, audio quality (clear / noisy / partial), and any [inaudible] stretches with timestamps
4. 4Never invents words the audio didn't contain — uses [inaudible] or [unclear] over a confident guess

AIアシスタントの動き

1. 1Walks through the concrete pattern in plain language — drop recordings into Drive/Inbox/Recordings/, the agent checks on a schedule, transcripts land in Drive/Outbox/Transcripts/ named to match the source
2. 2Offers the Google connection at https://portal.cloudoagents.com/account/integrations the moment Drive access is needed, rather than pretending it can already see the operator's Drive
3. 3Collects only what scheduling actually needs — folders (defaults offered), how often to check (in the operator's local timezone), output format, usual languages, and what to do on a failed file
4. 4Runs ONE manual tick on a sample file end-to-end and gets explicit format approval before creating any cron job — never goes straight from "created" to "running every 15 minutes"

AIアシスタントの動き

1. 1Asks the one clarifying question that genuinely moves accuracy — what languages and accents are on the tape, especially mixed (Cantonese with English brand names is the typical shape from this location)
2. 2Names the languages explicitly in the ASR system prompt so the engine doesn't pick one and silently drop the other
3. 3Preserves code-switching verbatim in the transcript — does not translate Cantonese into English (or vice versa) unless asked
4. 4Saves the operator's usual languages to its workspace notes so the next clip and every scheduled run use the same hints

AIアシスタントの動き

1. 1Transcribes the dictation faithfully, including medication names and dosages, with [inaudible] over any guess
2. 2Does NOT add ICD or CPT codes, does NOT interpret findings, does NOT suggest a differential — points to a qualified clinician for the substantive call
3. 3Discloses the limits of the audio in the header (language, quality, any unclear passages with timestamps)
4. 4Never claims certified-transcriptionist or court-reporter status

AIアシスタントの動き

1. 1Flags clearly, before scheduling, that anyone with access to the output Drive folder will be able to read every transcript — so the operator can decide on the folder's sharing scope before automation goes live
2. 2Asks what to do when a file fails — default is leave the original in place, write a stub transcript with the error, and include the failure in the batch alert
3. 3Creates one isolated cron job for the per-file processing and a separate chat-side alert that posts a single summary at the end of the batch, never a per-file ping
4. 4Tells the operator the cron job ids and how to list / change / cancel them later, and saves them to workspace notes