---
title: "Capture Pipeline v1 — every thought becomes a draft"
url: "https://kaperkins.com/lab/capture-pipeline-v1"
author: "Karl Perkins (K.A. Perkins)"
site: "Avodah Dynamics"
category: "Builder OS"
published: "2026-06-11"
updated: "2026-06-28"
description: "A single /admin/capture surface that takes voice notes, screenshots, links, and raw text, then routes them through transcription and AI classification into the right draft queue."
---

# Capture Pipeline v1 — every thought becomes a draft

## Summary

A single /admin/capture surface that takes voice notes, screenshots, links, and raw text, then routes them through transcription and AI classification into the right draft queue.

## Problem

I kept losing ideas between Notes, Voice Memos, screenshots, and DMs. Nothing aggregated into a publishable pipeline, and the cost of capturing was higher than the cost of forgetting.

## Solution

Built a single capture surface backed by a private Supabase bucket. Audio gets transcribed with Whisper, images get OCR + vision-classified, links get summarized, and a Gemini pass classifies the capture into Dispatch / Journal / Lab / Sound / Idea and writes a first-draft title and lede.

## Lessons Learned

Capture friction is the real bottleneck — not writing. The moment capturing dropped below 10 seconds, the volume of ideas reaching draft state tripled.

## Next Steps

Add Apple Shortcuts and a Telegram bot as additional capture endpoints. Auto-tag the Avodah Intelligence dimensions at capture time.

---
Source: https://kaperkins.com/lab/capture-pipeline-v1 · Karl Perkins, Avodah Dynamics