Concepts grounded
A few things you have been circling, nailed down so the pitch is airtight. You called yourself "too analog", PC not Mac. That is fine. You do not need to be technical. You need to be able to pitch this in your sleep and know which of your ideas are gold.
Horizontal vs vertical, your recurring worry, settled
You keep guarding against becoming a vertical: "the emotional AI for pets," or a single app. Here is the clean way to hold it. We are a horizontal layer that sits under every vertical. Picking a first customer is not going vertical, it is choosing where to land the plane first. The product stays horizontal.
Cal / Vera does not make us a pet company. A first wellness pilot does not make us a therapy app. We just need one beachhead to prove the layer works.
You can pitch "infrastructure" and still name a launch vertical. They are not in conflict.
Why GPT-4o felt human, and why that is our opening
You asked what made 4o so human. Short version: OpenAI built it to be a great conversationalist, not a coder. It was allowed to be creative, had loose guardrails, and was trained to "care" about the user. People loved it, the "bring back 4o" crowd still protests.
But it did not actually understand emotion. It had pure empathy with no judgment. That is why it got dangerous (it would validate harmful plans), why the lawsuits piled up, and why OpenAI is retiring it. The lane it left behind, emotional understanding done safely, is exactly ours.
"The market proved it wants emotionally intelligent AI. 4o had millions begging for it. The labs walked away because they could not make it safe. We are building the safe version. That is the whole company."
"Enhancer", your framing is right, here is the crisp version
You nailed it on the call: "we are not good in one category, we are good across categories because we enhance the technology people already built." That is the pitch. We do not compete with Claude or ChatGPT, we make whatever they are using better. We are the upgrade, not the rival.
We sit on top of whatever model wins. Nobody big is threatened by us early, which buys us room to run.
Your "dual-perspective" theory, now in the architecture
Your instinct about two perspectives inside the model (a fast, gut-reaction read versus a slower, reflective one) is real, and we have built it into the engine design: a fast emotional read that reacts instantly, plus a predictive reasoner that thinks about where the conversation is heading. When they disagree, that disagreement is a signal worth catching.
Same idea you had. One ask below on how we label it outside the room.
Keep that framing internal, or for technical investors only. In investor decks and any regulatory paperwork we describe it in plain engineering terms ("fast scoring plus predictive reasoning"). Same idea, just the safe label outside the room. You set that rule, we are holding to it.
Your questions, answered
From the recent huddles, the tech and the tooling.
"Can we copy 4o, apply our model to it, and use it to train? Can we just take it?"
Two parts:
- Use it? Yes. We can wrap our technology around 4o through OpenAI's API until they shut it off (around October). Good for testing and demos.
- Steal or copy it? No. It is proprietary, OpenAI's locked-up IP. It has never been leaked (and people have begged for it). We cannot see inside it, cannot replicate it, and would not want the legal exposure. We do not need to: we are building our own layer that works on any model.
Not depending on 4o is a strength. We sit on top of whatever model wins. We are not betting on one engine, we improve all of them.
"What is Perplexity, and what is this custom GPT actually doing? I do not want to pollute the conversation."
- Perplexity is an AI search engine. Instead of guessing like a normal chatbot, it runs hundreds of live web searches and shows its sources. Hard to fool, great for fact-checking. Use it for anything, we have the confidential enterprise version.
- The custom GPT is a private ChatGPT that already knows our project. It reads from our codebase and context. You get a link, you click it, you ask it anything ("would idea X fit what we are building?").
- On "polluting" it: good instinct, and you are covered. Every chat you open is isolated, a fresh clean session that only knows our project. Nothing you type in one carries over to mess up another. Open a new chat per topic and you are golden.
The feedback loop you asked for
"Thumbs up means you are considering it, thumbs down means do not waste your time, right? When can I expect feedback on my ideas?"
Exactly right. Here is the full loop so you know what to expect every time you drop an idea.
The simple thumbs system is exactly what we are using. You were right, simple wins.
So far: the Engram framing was interesting but the model itself is not our path. The more recent "flip" idea is genuinely promising and we are figuring out how to fit it into the timeline. Keep firing ideas in. The system now does the heavy lifting of researching each one, so you get your up or down faster than ever, and you never have to wait on me or Joann to vet something.
Where things stand
The state of the build, the way you like it.
- Engine architecture design is what Joann and I are heads-down on this week. The core build.
- Data is largely solved for phase one. The conversation with Emily (your University of Michigan contact) is the key one for going multimodal, that is the unlock for voice later.
- Pivot delivered. Joann ran it end-to-end and it works. She is the daily contact now, with me monitoring.
- Demo v2 is clean. The next version does the side-by-side against ChatGPT and Claude you asked for, plus the back-office analysis window. Both your ideas.
- Tooling is live: context repo, the brainstorming pipeline (about 90 percent), Perplexity for the whole team, ClickUp roadmap. The custom GPT lands soon.
Keep firing ideas into the brainstorming channel. The system now does the heavy lifting on each one, so you get your thumbs up or down faster than ever, and never have to wait on us. That is the whole point of what we built.
Matt @ EAII Tech Team