← All guides & prompts
Starter pack 02

Build your own offline dictation app

A practical build brief for your coding agent, with the failure modes and tests that turn a speech-model demo into a useful daily tool.

Windows · Version 1.0 · 6 November 2026 · 10-page PDF

Download PDF
  1. INPUTYour voice
  2. ON YOUR COMPUTERBark 3.0 Clean speech model
  3. OUTPUTText at your cursor

What it teaches

  • Speech recognition
  • Local inference
  • Desktop packaging

Build prompt

Build a personal English dictation app for my Windows PC in this project folder. Explain each handoff in plain English and do the implementation. Work in stages; verify each stage before adding features.

GOAL
A quick everyday tool: record speech, transcribe locally, insert the words where I was typing, and keep a short recoverable history.
Download text

Repair prompt

The failed test is [test ID].
I did: [exact steps].
Expected: [result].
Actual: [result and exact error].
Inspect the relevant log and the installed build. Identify the cause, make the smallest repair, and rerun this test plus the behaviour the change can affect.
Download text

Slow-performance prompt

Measure model loading and transcription separately on this PC. Use three saved recordings of different lengths. Record CPU, RAM, audio duration and elapsed time. Check that the model stays loaded between takes. Identify the biggest delay and fix one cause at a time. Keep output quality visible and compare the same recordings after each change.
Download text

How to use

1. Create a new project folder. 2. Open a coding agent that can run commands on your computer (for example Codex or Claude Code). 3. Paste the build prompt as one message. 4. Run the acceptance tests in the PDF before you rely on the tool.

Behind the build

Read how I made ToluWhispr