I cant text my friends without sounding cold so i found a way to make convo cues

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend

What I Built

I built properly, a local first convo buffer.
I cant talk to my friends properly, not because i dont care,but mostly cauz i cant think of anything in the moment.

Via properly, i can take a 60sec voice note of my day and get a cue card for use in convos

Code

GitHub Repository: https://github.com/Clumsyoof/properly

Built with:

• Streamlit(UI)
• OpenAI Whisper(Local speech to text inference running on CPU) -Gemma 2 (gemma2:2b) via Ollama(Local open-weight LLM inference)
• Local JSON Store

How I Built It

1. Audio is recorded in-browser via Streamlit's native audio buffer and handed to a local instance of openai-whisper.

2. The transcript is parsed into an episodic timeline stored strictly in a local JSON structure.

3. Ollama serves Google's open-weight gemma2:2b. Using focused system prompts.

Why Does Open Innovation Matter?

A closed api needs telementry and zero privacy, something which I NEED if i am talking about my day in a 60 sec slot.