Case study · 2025
IdeaVault: a voice message becomes a structured idea
A Telegram bot takes a voice message and returns the idea developed in layers — core idea, breakdown, mind map, SCAMPER and a review from opposing angles — in the same chat it arrived in.
- Role
- Freelance, end to end: workflow design and prompting
- Duration
- 2025
- Client
- Personal product, built to be usable by other people too
The problem
Ideas arrive while you are in the middle of something else. Opening a tool, typing the idea out and organising its details is enough friction for part of it to disappear — or for it never to be written down at all. The goal was to make capture fast enough that nothing is lost, without asking anyone to type or structure the idea first.
Constraints
- Capture had to be as fast and frictionless as it could possibly be.
- No form to fill in and no manual structuring before saving.
- A voice message had to work as the primary input.
- Processing and structuring had to happen automatically.
- It had to be usable by other people, not only by its author.
Decisions that shaped it
Voice instead of a form
Rather than designing a capture form, the Telegram bot is the entry point. You describe the idea the way it arrived in your head and send the voice message. From that moment the workflow does the rest, which is the only version of this that survives being used while driving or halfway through another task.
Turning a raw idea into something usable
The goal was never transcription. After the message arrives the system develops the idea in layers: a title and a date, the raw idea, the core idea, a breakdown, a mind map, a SCAMPER pass, a review from business, technical, economic and opposing angles, and related ideas. That shortens the distance between a thought that just appeared and an idea that can actually be examined.
The whole process automated in n8n
The entire process is one n8n workflow, so none of these steps is performed by hand. The bot receives the input, the content is processed inside the workflow, and the finished output comes back through the same bot — there is no second tool to open and no place for the idea to sit half-processed.
GPT for the analysis layer
OpenAI's GPT does the analysis, so the system does not merely record what was said. It extracts the structure of the idea and develops it from several perspectives at once — including the one arguing against it, which is the perspective a person is least likely to write down about their own idea.
What shipped
- Telegram bot as the single entry point for capturing an idea
- Voice message input and voice-to-text
- Automatic extraction of the idea's structure
- Idea title and date generation
- Core idea extraction
- Idea breakdown and analysis
- Mind map generation
- SCAMPER analysis
- Multi-perspective review and related ideas
- The complete n8n workflow, delivering the final output through the same bot
Have something similar in front of you?
Tell me the constraint you are working under and I will tell you whether this approach applies.