thought partner

an ai voice companion for work: reflect, brainstorm, and dictate notes we send a short email summary with takeaways and todo's after each chat

Comment

Inspiration

I've loved chatgpt voice since it was released. One evening I had a voice conversation that the transcript of filled 52 single spaced pages in a google doc. It's amazing, and way better than text.

But I've been frustrated by a few aspects of chatgpt voice and pi which make them not very useful for longer monologues of brainstorming and idea exploration -- (1) it interrupts me while I'm speaking, breaking my flow, and (2) it mishears words I say (e.g. "Figma" -> "fig mom" or "Zod" -> "cod") and I have no way to correct the transcript.

Also the new flirty voices are a little uncanny for me.

So I created thought partner to create an experience perfect for what I need.

What it does

Thought partner is a voice companion for productive idea exploration -- with thought partner I can talk at length about an idea I'm exploring, a concept I'm uncertain about, or just talk about my day. When (and only when) I'm ready, thought partner will respond with a thoughtful and caring summary of what I said and some ideas to consider.

After a conversation, thought partner will send me an email with the key points, takeaways, and TODO's from the conversation.

Unlike chatgpt voice or Pi, thought partner will only respond when I ask -- either with a button press or when I say a "go" word (currently "over", "go" or "echo").

Also, I can see and correct the transcript of what I'm saying before thought partner responds with voice -- this helps with proper nouns, weird words, etc. and greatly improves the response quality.

How we built it

forked the deepgram conversational demo and rebuilt to target the desired ux.

Challenges we ran into

websockets are annoying! and tons of frontend bugs.

Accomplishments that we're proud of

it works!

What we learned

I wanted to test if the "only respond when I ask you to" experience would be better than the chatgpt voice "eager-to-respond" experience would be worth the tradeoffs.

The tradeoffs are real -- it won't respond automatically within 10 seconds. I have to say a "go" word or press a button. But this is worth it for me, for long form exploration where I might have long pauses.

What's next for thought partner

an app! Also long term memory.

Built With

  • deepgram
  • openai
  • renditioncreate.com
  • vercel
Share this project:

Updates