This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend
What I Built
I built a local Table Topics practice partner for all my fellow club members (Rishi specifically) to prepare for the upcoming contest season. It give you a Random topic for your impromptu speaking, speak your answer, get a ballot-style scorecard and one concrete recommendation. Open models only, nothing leaves your machine.
Ballot weights: Content 55, Delivery 30, Language 15. Criteria wording is a paraphrase, not official Toastmasters text. Delivery is judged from the transcript only.
Demo
Demo Run Output
Code
Here is the documented code
How I Built It
Everything runs locally through Ollama. An open-weight Gemma model generates a fresh Table Topic and later scores the transcript against the ballot weights (Content 55 / Delivery 30 / Language 15). Open-source Whisper (faster-whisper) handles transcription. No API keys, no cloud calls - a speech you're practicing for a contest stays on your machine, which matters when the material is personal.
Why Does Open Innovation Matter?
A closed API would make every practice session someone else's data and someone else's meter running. Open models mean any Toastmasters club anywhere can run this offline and free - and because the prompts are plain text, a club can tune the scoring to its own ballot without asking anyone's permission.
My Agent Session
Prize Categories
Best Use of Gemma.
Friend's response
Rishi : "Bro, Integrate this with whatsapp! it would make it go viral across the world! If I skip the initial setup, its a gem after that! "
Top comments (0)