Rel.ai Filters & Prompts (with David)

# Transcript: 2026-04-28

> 3 time blocks from 1:42 PM to 2:09 PM

---

### Catching up about Rel.ai filters
**1:42 PM - 1:55 PM PDT** | *casual*

**Microphone:**
Hello. How's it going?

We talked about the filters. You remember when we started first on Raleighi and the prompt, and it was so important that it worked? Now with the filters, but the problem is he doesn't have my strong, very strong relationships. And I was like, so I showed him how that works.

Oh yeah, that's really long. Adding to that, yeah, Alan C. Mueller. It seems to... can't ignore fixed. But when you're starting on a fast follow-up, I said, I don't think this will be a big deal. I'll talk to you about it, but it shouldn't be a problem. And the idea turned on very strong and strong.

And so we have to be careful, obviously, because I don't think we want that to be an AI statement. I think if they do this, say this, and like they can either write or they can filter right from the freaking files. That's off for some reason, but I did the review one earlier. Like, this is already a vast improvement. Let's solve that. And then the second really sounds good to me.

Relationship leads and introductions are useless by themselves. Let me look at my typical companies. The encapsulation of a target company and then all of your orchestrations for that deal team, relationship leads and so forth should be a page. You should be able to, on your phone, swipe back and forth through the deal teams.

I was also describing this way that, like, if on the spectrum of assistance that we provide, relationship leads are the closest thing to a guarantee. Actually, that's not true. Offers for help are probably the closest thing to a guarantee for an intro. Relationship leads are next. And then there's matches or there's like, hey, you live in Detroit—should be focusing on that.

I think that all makes sense to me. I assume he has some basic heuristics to generate the suggested prompts, and maybe it gets massaged by an LLM, but I think we'll probably get very far with it.

I would suggest we try to stay away from using an LLM for this at first, purely because I think it'll be slow and I think users will be clicking it a lot. There could be some costs involved. So keeping it simple, I think a lot of this could even be done on the phone.

Just for the proper separation of concerns, build something on the back end. It might be useful in a bit. Make sure everything's ready for the prompt suggestion work.

All right. Thank you. I'll talk to you later.

**System Audio:**
Hello. How are you?

Do you remember when we started first talking about Rel.ai and the prompt? It was so important that the prompt—you never had to use filters. You could just ask what you wanted.

I had a long conversation with David. We were catching up about general stuff, and we talked about the filters.

Oh, just wait. No, I'm just kidding. Okay.

So I had a long time, now we're adding filters. And he's like, it's just weird to be able to find a little bug. Look at that. No, but look, it has plus... It's fine. It's a bug. We'll get it fixed.

Are they going to know what to type? He said, I don't think that's the page replacing the example questions. You know, if I start—if I go to relationship strengths. We want it to be an AI prompt generator, I think. That's like, you know, if they're toward companies, say this, if they do this, say this, and you know, combine them. But I don't think that should be that big a deal. And so we have to be careful, obviously, because I don't think we want that to be an AI statement generator.

I just want to know what my strong, very strong relationships were. And I showed him how that works. And he's like, okay, but you know, maybe senior folks or something. I'm like, that works too. I'm wrong.

You've also built a bunch—60 freaking files—with team introductions. Like, I want to get rid of the relationship leads and introductions dropdowns. As I explained to David, my view is that the encapsulation of a target company and all of your orchestrations for that deal—you know, for that deal team relationship leads and so forth—should be a page. And you go look at my relationship leads. What they do is they go, "Hey, let me go look at Postman. Let me go look at my partner companies." So deal teams, which are relationship leads and introductions, are useless by themselves. Like, they're useful, but without any context, it's just a page of these things. You don't know what the fuck you're doing. Nobody's going to be able to, on your phone, you know, swipe back and forth through the deal teams.

Well, that was funny. I wasn't even on that. Just popped over to something else. Yeah.

So I was also describing it this way: on the spectrum of what we provide, relationship leads are the closest thing to a guarantee. Actually, that's not true. Offers for help are probably the closest, and probably the least likely—not even an intro or a match, a chance, which is, you know, somebody who lives in the same town. That's the spectrum of potential assisting toward a guarantee for an intro.

Relationship leads are next. And then there's matches. Or there's like, "Hey, you live in the same town." And that's what the deal thing should be focusing on.

I would suggest we try to stay away from using an LLM for this at first, purely because I think it'll be slow. And I think they'll be clicking in the chat. It could have costs and stuff there. So keeping it simple—nothing hard-coded, to be clear. But honestly, I bet you can even build it on the phone.

Okay, yeah. And you could build off that PR, Relia. Thank you. Okay. Yeah. Yep, bye.

---

<details>
<summary>Background Noise (2 blocks)</summary>

### Brief unintelligible fragment
**1:59 PM - 1:59 PM PDT** | *background-noise*

**Microphone:**
. Okay.

### Unclear parking lot conversation
**2:07 PM - 2:09 PM PDT** | *background-noise*

**Microphone:**
But actually, how do I look? How do I clarify? We are. If they had it, they weren't. So I don't know how they were there.

The parking lot. He pushed himself. That sounds way more productive.

Girl? I'll give you something, don't worry. Give it to you. Nasty app.

</details>

The day centered on a detailed Rel.ai product discussion with David, revisiting the original vision that users should never need filters (just natural language prompts) versus the current reality of adding filter functionality. A key architectural decision was made: prompt suggestions should be generated without an LLM initially to avoid latency and cost, using simple heuristics that could even run client-side. The conversation also surfaced a significant UX rethinking — relationship leads and introductions are useless in isolation and should be encapsulated as pages per target company (deal teams), swipeable on mobile. A spectrum of assistance value was defined: offers for help > relationship leads > matches > proximity-based suggestions.