case://study-015
AI / ML
Project 015

EchoScribe

EchoScribe automates the meeting-to-Slack-digest workflow. Pull audio from a recording source, transcribe via Whisper, run a multi-step summarisation chain (decisions, action items, sentiment, follow-ups), and post a structured digest to Slack, Markdown, or JSON. Ships as a Python CLI, a FastAPI server, and a directory watcher — one OpenAI key powers all three. Docker image on GHCR.

CLI · API · Watcher
Surfaces
Slack / MD / JSON
Outputs
GHCR
Container
AI / ML
Category
The Problem

What Wasn't Working

Every meeting has 30 minutes of decisions and 5 hours of context. Good written summaries take humans 20 minutes to compose — and they always wait until morning, after the team has already moved on.

The Solution

How I Fixed It

A pipeline that runs the moment the recording ends. Whisper for transcription, a multi-step GPT-4o chain that extracts decisions / action items / follow-ups into a structured JSON schema, then renders Slack-friendly markdown. Ships as a CLI, a FastAPI server, and a directory watcher — pick the surface that fits the team.

Stack

Technologies Used

Python 3.10+
OpenAI Whisper
GPT-4o
FastAPI
Docker
Slack SDK
Results

Key Outcomes

End-to-end: audio file → Slack message in ~3 minutes
Structured-output prompts (no markdown drift across runs)
Three deployment surfaces from one codebase (CLI, server, watcher)
Container image published on GHCR for one-line ops deploys

Want something like this?

Let's build it. I ship fast and I ship clean.