Here's the thing. I want a personalized AI. Full stop.
There's no arguing back and forth.
Tech is here someone needs to put it together.
I'm so busy with other things, but I really want to see this thing come through and happy to work on it.
To me, its insane Apple didn't figure it out.
The goal: Build a truly personalized AI.
- Single threaded conversation with an index about everything.
- Bot should be accessible via WhatsApp, Insta, Email, or a phone call through an app (WebRTC connection).
(https://botpress.com/ is a really good starting point).
- Operator at the back that has login access to almost all tools I use, but critical actions must have HITL.
- Bot should be accessible via a call on the app or Apple Watch.
https://sesame.com/ type model and this is very doable with https://docs.pipecat.ai especially their new release that allows for long running function calls.
Speech detection (VAD) is still a problem, but I think its more of systems engineering problem than ML problem. Interruptions need to feel smooth.
But this is not a blocker. What exists is good enough.
- It can should be able to digest images, voice notes, etc.
- everything should fall into a single personal index (vector db).
One of the things could be, sharing 4 amazon links of some books I want to read and sending those links over WhatsApp to this agent.
It finds the PDFs for the books from https://libgen.is and indexes it.
I phone call the AI and I can have an intelligent conversation about the subject matter with my AI about the topic.
I do this a lot with xAI's Grok and its a surprisingly effective tool.
I give zero fucks about issues like piracy at the moment.
I want to later add more capable agents as tools to this AI.
There's no arguing back and forth.
Tech is here someone needs to put it together.
I'm so busy with other things, but I really want to see this thing come through and happy to work on it.
To me, its insane Apple didn't figure it out.
The goal: Build a truly personalized AI.
- Single threaded conversation with an index about everything.
- Bot should be accessible via WhatsApp, Insta, Email, or a phone call through an app (WebRTC connection).
(https://botpress.com/ is a really good starting point).
- Operator at the back that has login access to almost all tools I use, but critical actions must have HITL.
- Bot should be accessible via a call on the app or Apple Watch.
https://sesame.com/ type model and this is very doable with https://docs.pipecat.ai especially their new release that allows for long running function calls.
Speech detection (VAD) is still a problem, but I think its more of systems engineering problem than ML problem. Interruptions need to feel smooth.
But this is not a blocker. What exists is good enough.
- It can should be able to digest images, voice notes, etc.
- everything should fall into a single personal index (vector db).
One of the things could be, sharing 4 amazon links of some books I want to read and sending those links over WhatsApp to this agent.
It finds the PDFs for the books from https://libgen.is and indexes it.
I phone call the AI and I can have an intelligent conversation about the subject matter with my AI about the topic.
I do this a lot with xAI's Grok and its a surprisingly effective tool.
I give zero fucks about issues like piracy at the moment.
I want to later add more capable agents as tools to this AI.