Daily Log Automation — Voice-to-Digest Pipeline by Kenny DunnDaily Log Automation — Voice-to-Digest Pipeline by Kenny Dunn

Daily Log Automation — Voice-to-Digest Pipeline

Kenny Dunn

Kenny Dunn

The Problem

Manual end-of-day logging gets skipped exactly when it matters most — under deadline pressure, the decisions and blockers worth capturing are the ones that never get written down. Most "just write it in a doc" habits die within a week.

The Build

I designed and shipped an n8n automation that removes the friction entirely. It runs on two paths:
Capture path (event-driven):
A Telegram bot listens for incoming messages — text or voice — from an authorized user only (chat ID verified before anything else runs)
Voice messages are pulled via the Telegram file API and transcribed through Google Gemini 2.5 Flash
A custom JavaScript parser normalizes both text and transcribed voice into one schema, extracting structured fields (date, category, status, decisions, next steps, active projects, tags) out of freeform language
The parsed entry is appended to a Google Sheet using the message ID as an idempotency key — no duplicate rows, even on retries
A branded HTML confirmation email fires as a receipt
Digest path (scheduled):
A nightly trigger pulls the day's log entries
An LLM call generates a JSON payload: AI-written summary, localized weather block, and relevant tech/creative news headlines
A second parser handles that JSON — with a regex fallback in case the model doesn't return clean JSON (fenced code blocks, malformed responses, etc. all handled)
The final digest is appended into a running Google Doc formatted like a daily newspaper: heading, status, decisions, next steps, risks/blockers, news
Error handling: a dedicated error trigger catches failures from any branch and logs them to a separate sheet with execution ID and workflow name for fast triage — this doesn't just fail silently.

Why It Matters

This isn't a toy webhook demo — it's a production pattern: multi-branch conditional routing, AI transcription, LLM-based structuring with a parsing fallback, idempotent writes, and dedicated error logging. The same architecture generalizes to intake forms, support ticket triage, or any workflow where unstructured human input needs to become structured, actionable data.

Result

Zero missed log entries since deployment. Capture takes under 10 seconds from a phone. The voice-note path specifically solves the "too tired to type after a long day" failure mode — you talk, it's structured and filed automatically.
Watch the full demo (voice note → structured Sheet row → digest) here: https://drive.google.com/file/d/154F48txJLBQGXj7PUkpcT9qsGelSa5MV/view?usp=drivesdk
Like this project

Posted Jul 21, 2026

n8n workflow that turns a voice memo into a structured, AI-summarized daily business log — zero manual entry.