Automated speaking coach for Airtable practice logs

Build me an agent workflow that turns our Airtable student practice log into an automated speaking coach with just-in-time teacher alerts.

Trigger: poll Airtable for a New Record in the practice-log table. Each row represents a student submission and contains the student's name, a target sentence or prompt (may be blank for free-form practice), and an attached audio recording. The base ID, table ID, and the specific field names for student, target text, and audio attachment should all be configurable.

When a new record fires, the agent should do the following in order:

1) Read the audio URL from the attachment field on the row.

2) Pick the right Speechace scoring method based on the row. If the target text field has a value, call Speechace's Score Text (Pronunciation, Fluency, Intonation) operation with that target text plus the audio URL. If the target text field is empty, call Speechace's Score Speech (Open-ended, Relevance, Language Detection) operation with just the audio URL. The dialect (default en-us) should be a configurable input.

3) Extract the overall pronunciation score, the fluency score, and the standardized CEFR and IELTS bands from the Speechace response.

4) Use Airtable's Update Record operation on the same record ID to save the pronunciation score, fluency score, CEFR band, and IELTS band into their respective columns on the row. Do not touch any other fields. The target column names for these four outputs should be configurable inputs.

5) Decide whether to alert a teacher. Before deciding, use Airtable's List Records to look up the same student's most recent previous submission for a prior pronunciation score. Alert only when either (a) the new pronunciation score is below a configurable threshold (default 70), or (b) the new pronunciation score is lower than that student's most recent previous score. If neither condition is true, do not post anything (silent success is the default so teachers only get pinged on students who need help).

6) When alerting, use Slack Bot's Send a Message operation to post to a configurable channel. Compose a short, warm coaching message that names the student, states the pronunciation and fluency scores plus the CEFR or IELTS band, highlights the top one or two mispronounced words or fluency issues pulled from the Speechace response, and suggests one concrete drill (for example, a minimal-pair drill for a specific sound the student missed, or a slow-then-fast repetition drill for pacing).

Configurable inputs to expose: Airtable base and table for the practice log, field names for student / target text / audio attachment / output columns, Speechace dialect, pronunciation-score alert threshold (default 70), and the Slack channel to post coaching alerts into.

Use the Slack Bot integration (not user Slack) so alerts come from a shared workflow identity that keeps working even if a specific teacher leaves. Coaching message should be plain Slack mrkdwn, kept to a few short lines so it reads well on mobile.