Introduction: Vibe Coding Meets Product Thinking

I'm going to present product -oriented thinking more so than technical thinking, but it is also technical. What I'm going to talk about is vibe coding, analytics, and some different ways that I've come through a journey through doing that with agentic approaches and just trying to really marry what's feasible and possible with needs and solving problems.

So my background is kind of two tracks in a way. I have an analyst background, worked in analytics and data in fintech primarily for most of my career, but I also had a background for many decades as a creative. So screenwriting, filmmaking, music making as well.

Analytics in Unexpected Places

And so where I'm going to start this journey to kind of take you through to what I've just built like literally yesterday, is thinking about analytics in places where you might not expect them, right?

Vertical Microdramas and Structured Storytelling

So one of the things that I'm doing in my screenwriting work right now is I work on verticals. I don't know if you know what verticals are.

It's content for your phone, it's a 916 format, stories that you watch, you know, places like Drama Box and Real Short, these little dramas that pop up on your phone. So I've been writing those.

and as I've been doing that I found there's a really interesting intersection and you wouldn't expect that but in in storytelling and data analytics there's a huge convergence right now in this space so these dramas are structured even though they're video entertainment they're very very structured content you start with a hook you escalate the stakes of the story and then you have

a cliffhanger repeat repeat like about for about a minute and a half episodes so when you think When you think about short -form content, you think about these analytics, for me, I saw an opportunity.

Write Rocket IQ: An Engagement Model for Short-Form Scripts

The first thing that I built in that space is this app that you're seeing right here. It's called Write Rocket IQ. What it does is it takes a screenplay that's designed for this format and it analyzes that screenplay based on factors.

It's like a multi -factor model and this app is hunting. the model is, well, the analysis, right, is looking for certain heuristics. If you see a face slap, someone's going to lean in and watch, right? These are known behaviors.

So imagine this story that's being told. This one is one that I wrote called Dragon, The War of the Dragons Sworn, okay?

It's about a woman with a dragon. Somebody steals her dragon. She goes after it. There's a love story, you know, et cetera.

And so my question is, well, are my viewers going to watch this or not right so what i did was i just said well we know now from all of the research that's gone into these stories um what makes people watch and not watch and so i built this

tool to literally go beat by beat in this in this uh script that i've uploaded and look for the things and tell me back um where i need to focus my attention to get better engagement so it's an an engagement model for short form content. And that's literally what it's doing.

You could imagine in a one to two minute episode length, right? You're going to have 50 episodes. There's

a lot of data to go through, especially if you're a writer and you're creative trying to sort of picture the story, you're not necessarily thinking about data. So this is the bridge between that world and it serves the need.

So the need or the problem that I'm solving is for writers who want to make sure that their writing hits that bar of engagement particularly when the data folks don't really speak the same language as the writer team or the creatives right so there's this gap to fill and I thought well let's make a

tool because I speak writing and I speak data I'll make a tool so this has been really interesting got a lot of getting a lot of positive feedback in the I'm

in a pilot mode right now and then I realized well you know every platform wants their own secret sauce potentially right there's behaviors that are common but then also what if I'm doing true crime or I'm doing a whore instead of romance there might be different things

Making the Model Configurable (Multi-Tenant Tuning)

that are going to engage right so I said what I really need is I need this multi -tenant version where people can actually design their own adjustments to what they want to see in their content and maybe they're getting information from the audience right so then I made it

configurable so anyone who uses the the the tool can go in and configure their rewards and penalties what's getting you know analyzed so once you kind of tune you can tune your model however you want you can see a preview of what would what

impact would that have on your story how is it going to be analyzed and now you get hopefully a closer analysis on the next thing you analyze okay so you can And sort of see, oh, yeah, there's a learning loop being built here.

Closing the Loop: From Script to Production Video to Audience Metrics

Well, then it occurred to me, well, there's another problem to solve then, right? Because I've got a script and I've analyzed the script. But then what happens when you go to production?

What if the director spends three minutes focused on a coffee cup and the audience wanders away? Right? So there's a gap then between the writing and production.

So I'm not going to show that now because it's kind of a heavy lift. but so then I added analyzing video with the Gemini flash so we now can analyze video and see where you have gaps between the script and the production

and then the final question that came up was well how did it do though what about the metrics what about the audience actual results right places like YouTube and and apps can see where viewership scrolled away well let's now retrain so then the final piece is getting the analytics data load the analytics data data, and then going back to these configurations, you get recommendations on what to change.

So just kind of following a trail, effectively, of problem, how do I solve it, problem, how do I solve it, and just kind of iterating through, vibe coded. So yeah, vibe coded analytics, multi -tenant model.

AI Filmmaking and the Broader Conversation About AI

So I'm gonna take you through two more pieces of this story, because I just find it interesting. So I'm also an AI filmmaker, right, besides being a writer, and I went to a hackathon recently and spent some time making an AI video, but I'll tell you the prompt, right, so considering this work I've been doing in this background, the prompt was kind of like what do you want to see in the world from AI?

And so when my teammate and I were thinking about this, I started to get a little bit of a different picture. And I don't know if you guys have in your own personal lives people who are backlashing against AI.

I don't know if people in your social circles who say, I don't want it. Get it away from me, right? It's a very strong backlash that's happening.

And so what my creative partner and I said was like, well, look, we are surrounded by creative friends and associates. We want them into the world we know. We want a partnership.

We don't want this giant divide. And so we made a video that spoke to that, and I'm just going to put the mic down so you can see it's a very short video, but this is the content that we saw. Let me get the mic to go here. Let's restart it.

Hackathon Short Film: What Kind of AI World Do We Want?

We need to keep documenting this. AI turbocharged achievement award. We've optimized every parameter. He's in the top percentile.

This nanobot injection eliminated microplastics in my body a thousand times. know. Algorithms that feed you what you want to hear are a terrible way to get medical advice.

I'm billionaire Jim Robertson and I'm here to help the world. I'm going to give you this brand new computer if you can answer just one question. Are you ready?

Sure. What is the question? What is the best company in the world?

Robertson Labs. Robertson Labs cares about our world and you should too.

Still take breaks. I just don't see real people much. What do you think?

Nobody asked us if we wanted this and it doesn't feel safe. Nobody asked me about this.

They just said it's AI. Is this what we even want? We kind of need to talk about this.

Can we build a world where today's technology isn't tomorrow's landfill? How about AI that knows we don't all look the same? Because we don't.

You saw funny cat videos, and I love that. Can we maybe think a little bigger now? Bigger problems, not bigger cats.

How about AI that preserves culture? this thing it's more than data much more can ai build a world where kids are so curious they don't outsource their homework to ai enough with the deep fakes already can we start building

deep trust back we were promised flying cars i think we maybe need less cars overall oh yes look at this you could have walked i don't want ai to take everyone's jobs everyone needs money money to buy tickets when my movie comes out suppose you think ai is the answer to everything

actually i don't that's why we have to keep talking can we build a world where we don't have to shout just to get heard we should try for sure is um so that is where i

Building Analytics for Good: Go Life Green

I jumped off for the next thing I needed to build. And what happened was, you know, I've been building this, you know, analytics tool, been doing the AI filmmaking,

So, I mean, that video is just sort of a statement about trying to be inclusive, right? is making that video and having conversations made me turn the lens in a little bit.

I said, well, what am I doing, right? I'm out here building analytics tools at 3 a .m. with vibe coding, and I'm, you know, I'm making AI videos, and I'm not really living that same, you know, philosophy, right?

Well, how do I make the world better in some way, shape, or form with what I'm doing? And so I started to try to add that in.

And so I made, you know, a new app. I made this one that's called Go Life Green. And so this is my attempt to sort of use my make an analytics app for good. rather than just for, you know, engagement metrics and short form content, et cetera.

The Personal Problem: Reducing Chemicals and Microplastics

So one of the things in my life that I am trying to do with not much success is try to live healthier with fewer chemicals and microplastics all around me, right? And it is really hard. I don't know if you've observed or watched any documentaries.

Like, it is really hard. It's everywhere. So I said, well, maybe other people like me could benefit if I build something to help with that.

And so I did, and so I will, I'm just gonna pull up a couple slides on it because I think it's easier to understand.

So I built a product that helps you track, you know, all of the substances, chemicals, microplastics in your environment, and helps you, you know, nudge to get better over time.

You know, not, you're never gonna get perfect, never gonna eliminate them all, but I'm trying to do that for myself, and I thought that, well, let me use my skills for that.

How the App Works: Regulated Ingredient Lists + Multi-Agent Analysis

that and so one of the things that I have done is realize like I took a trip recently to Europe and Europe has a much stronger regulatory system to track these things and and look product by product so I said well let me grab that list of ingredients that they have banned or regulated

so basically going out to the sources it's very hard to do on your own I mean of course you could build a Claude skill or you know custom GPT but if you don't want to bother to do that I've done it for us so you know I built that to analyze the ingredients based on just looking at an

ingredient list on a product because that's literally what I was doing in this in the in the shopping shopping mode I'm looking at them I'm looking around my house the cleaners the

toothpaste etc and so I did that with a multi -agent model so first thing it does is it looks at an ingredient list. Then it says, well, you know, using something on your skin versus kind of, you know, passing by it might be a different type of context.

There's usage context that started to matter. I realized I built an agent to analyze that. The exposure path, how frequent

you're using it. And then LLM is judged to say, okay, these agents are bringing back all this information. We need to make sure it's okay.

And then finally to do a nice write -up on what that that risk is that you're experiencing. So I did that. Um, and this, I did it in about a day with

vibe coding, um, because it's just, it's kind of that easy. So yeah. Um, the schema, the edge functions, functions and super base, I like lovable.

Um, that's, you know, my, my tool of choice. Um, but yeah, you know, the auth is already there for you, super base backend, um, and easy to spin up these agents.

Wrap-Up: A Journey of Iterative Problem Solving

So yeah, that was my journey. I just thought it was interesting to talk about different ways to look at analytics different ways to kind of bring all the vibe coding or you know building that you might be doing into your world in different parts of your life

Q&A: Accuracy, Testing, Distribution, and Market Fit

that's it yeah on that last app i was just curious it's kind of a question i have about ai projects in general really is like how do you know that those results are correct and not just sort of like are you looking at the sources of the information or is it sort of because we know ai is kind of like this mirror of published humanity so how do you know it's sorting out out what you want from the hearsay.

Yeah.

Deterministic Data vs. LLM Judgment (Avoiding Hallucinations)

So I think there's a time and place to be deterministic and there's a time and place to be non -deterministic.

So the deterministic part is I literally have taken the database that the EU uses and stored it, right? So I have a very structured database that the agent will go and look through.

And then the matching exercise is where like the natural language comes in, right?

So if I've got an ingredient list and the ingredient list is a certain order, right? It's going through that ingredient list, looking at the database. So very deterministic, very, you know, non -AI -ish zone of work in that space.

Video Analysis Inputs: Transcript + Visual Timing

For the first one, when you showed about the videos and then understanding where to put the emotions in place, I guess, are you using transcripts from the video or is it the images itself that you're loading to the AI or like LLM?

Yeah, so it's both video and transcript. So, and the reason for that is, and I know this as a writer, right? When I upload a script, there are action lines that can be tons of information. Someone gives someone a dirty look, or they gasp.

There's a lot of things which are not explicitly in the transcript, maybe the way that I would expect them to be. And also timing. There's a lot of study on how many seconds it takes to engage someone.

So I don't want to miss, and that's part of the analysis from script to production, direction, is where do you have gaps or where do you have big spaces, so you're sort of kind of rationalizing from something by the script you might expect to take, you know, a couple of seconds, but in the video, it's taken a long time for that shot, and so you're kind of basically noting that, so it doesn't sort of corrupt the analysis.

Gotcha. Have you tried the accessibility transcript tool, which actually says somebody's looking here, you know, like it's more rich?

Yeah. Yeah, I think that that is, I think that that level is captured. I guess it's maybe a preference around how specific the writing describes it, right? But I think you're right. I mean, I think it's pretty close and it's probably just like a, I don't know, it's marginal difference probably. Yeah.

Sorry, last one on this.

so here how do you know like what ensures that when we say like say the ingredients which are healthy what if they are wrong like you know what happens then like say the llm got it wrong one of the agents was like hallucinating um so how do you how are you protecting against that

yeah so i have llm as judge so i have a separate agent that's all it's only job is to is to check for like kind of sanity check the results so in addition to my you know kind of you know human check sort of more like a spot check of results I also have one agent that is just trying to deliver accuracy yeah just piggybacking

on what he said would it make more sense to have like a golden test set of just you know images in different conditions different cameras just to see where it it may fail and where it doesn't, right? Any changes you make to your software,

you can always run it on the golden test set just to see what is the actual accuracy instead of depending on an LLM judge.

Yeah, no, I think that's a great idea. I mean, I think that's one of the test suites that would be kind of ideal to kind of add to the mix, yeah.

Hey, hi.

I've been using Yooka app for the, you know, scanning the product and finding all the ingredients. Yooka. Yooka, oh yeah, Yooka, huh? So how can it be different? This application would be different from that.

Well, mostly I would say the breadth of things being checked. So I'm looking at, you know, anything that's sort of banned in the EU, regulatory sort of like criteria, as well as the packaging that the product comes in, the context you use it, the frequency you use it.

and then I'd say probably you know the only other real difference I think that's meaningful is the ability to track so I've got in my mind sort of like my other apps that I use are like my fitness pal and things like that where I'm like I just want I just want to personally track am I doing better now I feel like I'm doing better am I doing any better so that that piece of the you know know, behavioral tracking, I think I'm doing a pretty good job on that. So yeah.

Thank you.

Vibe Coding on Mobile: Web App Hack vs. Native Testing

My question is about vibe coding. I have never developed an app in Android. And so my question is actually very primitive.

If I use vibe coding to develop an app, which I'm trying right now, do I still need to worry about like how you work on the different versions of Android, different hardware, and all this kind of thing. Do I need to test them out in all these different scenarios?

Yes. And that is a bridge I have not crossed yet. And I will tell you my hack, which is a complete hack.

When you sign up, let me see if I can just sign in quickly and see if I can get it to it come just to happen here so I can show you. The answer is yes, and I have not done it yet.

But this is a three -step wizard guide to tell you to put the web app icon on your home screen. Basically make a shortcut on your home screen. This is a web app. So I built it as a web app.

I did not build it as native Android or iOS. iOS, I literally just take my user and say, hey, step by step, make this a shortcut on your phone. That's it.

That's it. Until I get traction and I know people like it this way or that way, I'm not going to invest in trying to make it native mobile.

So basically you are telling people to put the app on your home screen? That will result So you can just click it like it, as if, you know, give people the sense it's on my phone because I scan my ingredients with my phone, right, the phone camera.

So all it does is it just walks you through a couple of steps. It says go to your web browser, to the website, right, and basically click add shortcut. It just, for people who don't necessarily think that way, it tells them what to do.

So it's just a hack until they get there. My question is about testing the app on the different hardwares, you know, Samsung, Pixel,

and on the different versions of Android. Yeah, for a web app, I'm not worried about that. For a mobile app, I'm worried about that.

For a web app, I don't know if you guys agree, but I don't see any issue with that. Yeah, it's a web app. Okay, got it.

All right.

Go-to-Market: Finding the First Users and Choosing a Persona

Hi, I love your tool project, and you are also sharing the experience that why you are doing it because you have the problem so you want to solve it

and I am also building my project but I found that so my question is how do you get your first 100 customer or user how do you to do the go -to -market step

and for example the first one I believe a lot of writer or person who trying to do the short video need it so bad and how do you talk to them

That is a great question. So the analytics app for the short form content, I get a lot of demand from writers. And I've really been pushing back on making that just opening the door to writers. And I may change my mind on that eventually.

But the overhead as a solopreneur, and that's on that app, to manage their questions. Writers are supremely non -technical. so I'm worried about the customer service element of dealing with writers so I have gone more b2b

so I'm actually promoting my app to producers and platforms who work with a pool of writers as a way to sort of reduce my overhead of dealing with some writer decided they could write they you know they pay their twenty dollars or something and now I've got you know dealing with them right so it's a yeah it's a challenge but they go to market I mean

this is you know it's a it's a big question and I don't know that there's one universal answer but for me it's been a lot about really who is the persona that I want to serve and that I that I can feasibly serve right because

you think about the platform and the configurations and uploading the videos like there's all there's a little bit of a white glove element to it that I you you know, I've been, uh, hesitant to open up just to every writer.

Thank you. Thank you for sharing. Yeah. Okay.

Unions, Microdramas, and Whether This Scales to Hollywood

So, uh, as you know, Hollywood's a union town. Um, most of the unions have been quite vehemently, uh, opposed to AI, uh, adoption.

Uh, most of the vertical micro drama productions have been non -union. I was just wondering whether that provides like a playground for you essentially

to develop the analytic application. Whether you think it might eventually cross over into Hollywood productions and whether you think that like vertical productions

are like a sufficient market to sustain new applications such as yours long term.

Yeah, I love that question, because it's changing, like, every day. I think yesterday's big news, I think it was yesterday, was A24, right, huge kind of prestige production house, got, I think, $75 million invested by Google for AI capabilities, to build out their AI capabilities. Now, they're saying it, I believe, don't quote me on this, I think they're saying mostly for previs of the pipeline, but still, that's huge.

Netflix is in micro dramas, I mean, there's just all of these intersections between AI, micro dramas, and traditional Hollywood that are happening now. And that's one of the reasons when I first started vibe coding that I actually got into to microdramas in the first place,

was I thought, well, microdramas move so fast. And just to give you like a sample, from start to finish of production is about, you know, it could be as little as a month and a half,

two months, the entire thing, 50 episodes, one and a half minute or so episodes. So it's like doing a whole feature from the writing to the production

in just like a couple months. They'll do, you know, shoots of, you know, six days. You know, it's really crazy, the speed and the output of micro dramas.

So my hypothesis was well this is the place we're gonna get AI first because you have to move so quickly that's the that's the AI zone and that's why I sort of started to angle my my work and thinking around that space and that is actually where you are seeing tremendous

number of AI productions just 100 % AI coming out. So I but I am I from where I sit, I am seeing a ton of intersection.

Again, even as someone like Netflix or TikTok, other places get into micro dramas, if they're going to keep up that speed, most likely it's going to be at least a hybrid AI. And then also, you can see the rise.

Here's a plug. If you're interested in the space as well, I would recommend Machine Cinema. I'm a volunteer with Machine Cinema

that gives free workshops on filmmaking using AI. They're in San Francisco. They're all over the world, but San Francisco, L .A. There are big conferences now in L .A.

just all about AI filmmaking. And the buzz that I hear is that the jobs are increasing. So first there was a fear about job loss, and now we're seeing job increase.

So I'm hopeful because obviously I'm here, I'm in AI, and I believe that they're going to co -exist. Filmmaking and AI is going to co -exist and give indie people like myself also more opportunities.

So yeah, that's my read.

Yeah.