Parker Rex
All videos
Parker Rex Daily

Why Everyone’s Talking About GCP + Gemini + Vertex AI

April 18, 2025

Watch on YouTube@parkerrex2

Parker digs into building a GCP-powered pipeline to turn daily videos into ready-to-publish assets, then sketches out the next wave of automation like thumbnails and deep-content research.

The pipeline in action#

  • OBS records long-form video; once encoding finishes, the file lands in a Google Cloud Storage bucket mounted on his Mac (GCS Fuse).
  • A Cloud Run function watches the bucket, detects whether the input is a daily or main video, and triggers Aonic for transcription and asset generation.
  • Aonic outputs:
  • Video MP4 and an audio-only file
  • An HTML file with: show notes, chapter markers, a long summary, a full transcript
  • Premiere Pro renders the RAW video for workflow throughput; the assets flow back into YouTube-ready content automatically.
  • All of this is driven by a small, Grock-scripted workflow that handles mounting, watching, and routing to YouTube with channel-specific data.

The stack and how it fits together#

  • Core platform: Google Cloud Platform (GCP)
  • Cloud Storage (buckets) for media and assets
  • Cloud Run to host the automation logic
  • GCS Fuse to mount large buckets locally
  • Vertex AI for prompts and structured outputs
  • Transcription and assets: Aonic (Whisper-based)
  • Local-to-cloud workflow concepts:
  • “Vercel-like” deployment via Cloud Run
  • Structured outputs from Vertex AI (JSON-like results)
  • How it looks in practice (snippets)
  • Cloud Run trigger (pseudo)
    # on new blob
    video_type = 'daily' if 'daily' in event.name else 'main'
    trigger_aonic_transcription(bucket, file, video_type)
  • Vertex AI structured-output prompt (example)
    {
      "task": "analyze_comments_and_transcripts",
      "transcript": "<full transcript>",
      "comments": ["..."],
      "output_style": "structured",
      "sections": ["summary", "chapters", "use_cases"]
    }
  • Why this matters: pricing power and scalability—GCP lets you push thousands of operations with minimal cost and predictable routing.

Current capabilities and workflow#

  • Outputs you get per asset set:
  • MP4 video
  • Audio-only file
  • HTML document with: show notes, chapter markers, short and long summaries, full transcript
  • Automation targets:
  • Channel-appropriate YouTube uploads with associated metadata
  • Thumbnail generation planned next
  • Core advantages:
  • End-to-end content generation from a single source file
  • Extensible pipeline that can incorporate new data (comments, sentiment, trending topics)

Roadmap and future ideas#

  • Thumbnail automation: generate high-quality thumbnails programmatically
  • Idea generation and validation:
  • Pull from comments and transcripts, perform sentiment analysis, extract open loops
  • Seed ideas into a content ideas sheet with tags (coding, AI, marketing automation, etc.)
  • Validate ideas using sentiment cues and trend signals
  • Deep research layer:
  • Use sentiment + transcript context to surface three practical, high-value use cases per audience (e.g., marketers, coders)
  • Build a “plus-up” context block to inform video planning and future scripts
  • Cross-service automation:
  • Extend the pipeline to account-based actions (e.g., other creators’ channels for research prompts)
  • Automate YouTube transcript chapter markers or other meta tasks via API

Learnings and practical tips#

  • GCP is more capable than it looks if you map the architecture first; it unlocks cheap scaling and rapid experimentation.
  • Mounting long-term storage locally (GCS Fuse) is a practical workaround for large media workflows.
  • Start small: a simple bucket-watch -> transcription -> HTML asset is enough to prove the value; you can layer complexity later.
  • Expect some early friction around credentials and secret management; use a dedicated secret store and proper access controls.
  • Tools like Grock/Cursor are useful for prototyping, but plan a clean architecture early to avoid version-control chaos.

Quick wins you can implement (copy-pasteable ideas)#

  • Set up a Cloud Run service that watches a bucket and triggers a transcription job whenever a new video lands.
  • Use Vertex AI with a structured-output prompt to extract chapters, summaries, and sentiment insights from transcripts and comments.
  • Mount your bucket locally with GCS Fuse to preview assets and streamline rendering in your local tools (e.g., Premiere Pro).
  • Start a simple pipeline to generate a YouTube-ready HTML description (transcript, chapters, summaries) and test uploading with a test channel.
Transcript

Let's work on the next task that is high priority. Cool. I don't have any credentials up. Hey there, I'm Parker Rex. I led tech for a startup that sold for 23 million bucks. After that, did Airbnb for music. It was stupid. Went all in on programming. And now I show you the strategy on my way to 100K in profit a month using AI. So, I've spent a lot of time thinking about how to use it. And this is a daily upload. I've just been having so much fun with Taskmaster. It's like all I can use. So, I've been yapping on this channel and it's been pretty boring lately to make these videos to be honest because I haven't been building that much stuff because I've been building the community and like putting all this energy into the community and I think it's kind of rolling now. So, I got back to building. So, usually I go through questions and answers, I do some news and then I do the builds. But, I'm just excited about the builds. So, let me just yap for a second. Okay, so I talked about this a long time ago, a whole month ago, where I was going to be making a system that takes these videos, not only from the daily channel, but also the main channel, and it would turn them into a bunch of different pieces of content. So that I'd have 10 pieces of content that were long form and then I'd have all of that go and turn into clips and turn into blog posts and turn into medium posts and turn into Pinterest pins and turn into Instagram carousels and LinkedIn posts and boom everywhere but keeping it high quality. And let me tell you what I did today. It's a lot of fun. So I originally made this and I was like cool. You record the long form boom you transcribe it. Duh. Then you export it to G Drive and use Make or N. Didn't do any of that. Instead, I went down the good old rabbit hole that is GCP, Google Cloud Platform, and it's been awesome. So, I was thinking, okay, how can I make this so that I have it go from, let's see, from camera to South Carolina, like a a computer that I rent in South Carolina from Google. How do I get this to encode the video straight onto a mounted bucket that has like a pabyte of access, which I'm never going to use that much, but the storage is super cheap. And I need it hosted there because I need this service called Aonic that does the whisper transcription and all this stuff like it creates these assets. You get like a HTML file with a summary and like mid mid form, long form, all this stuff. So I'm like, "All right, I'll just take a quick look see into GCP." And then I'm like, "Wait a second. I'm already in here. Let me let me see what this thing's capable of." And so I end up basically, let me just try to kind of draw it out. I end up basically doing it all in GCP with cloud functions. So what's happening now is long form gets recorded, right? This is on OBS. And I take that and then as soon as it's done encoding and moxing, if you know what that means, that goes straight into a bucket. And I can show you what that looks like. So, this is mounted to my desktop. It's called GCS fuse. So, if you ever need a lot of storage, this is one PB pabyte. Crazy. this structure because it's mounted here. When an item lands in here, I'm going to render it straight from Premiere Pro. So, if you made content, that's one of them. It's going to render straight from there into RAW, either daily, if it's a daily video, or raw main. And so, what's cool about that is the second that that goes in there, then it's being watched and it's being listened to. And so that goes in there and then boom, that gets kicked off to Google Cloud. And then Google Cloud allows you to write a little bit of code where let's see, let's do cloud run these buckets. So this bucket, you can think of this, I deleted a lot of stuff today. You can think of the bucket as just a fancy G drive, but it's connected to a bunch of other stuff. So now when I boot my computer, that pabyte of data is there for me to just throw stuff in. Really cool. So I have buckets, right? Said buckets a lot. And then what I do is I have this thing called Cloud Run. And Cloud Run has a function in it. And this function is listening for when stuff gets into it. So you haven't you see I haven't like run it that many times because I know that that part works but it's listening as soon as it gets that it determines is this a daily video or is this a main video and then it'll just shoot it off to Aonic and then Aonic does its thing and you can see it's doing its thing now on this video that's about to go out and what you get is a couple different things. You get an MP4 file. You get a audio only file. So this big boy, the 12 gig one that is the video obviously. Then you have this MP4 which is just audio. Then you have this HTML. So let's grab that. And this API that it has is actually like pretty dope. So let's grab this HTML. In this HTML you have the show notes. So the little chapter markers. You also have how it's like this oneliner about the thing, right? So, this is the transcript, the long summary, chapters, the brief summary, full transcript, and then if I went Yeah, that just like brings me up and down. So, that's one of the assets. And okay, cool, dude. You could have done that by drag and dropping. Yeah, but I'm thinking in terms of extensibility, it's kind of endless. like let's see if I can find the tab. But basically in here, this is just one little blob of code and I didn't write any of it. I had Grock write it and I worked on Grock for like two hours. Got it to do this and then the mounting and then now the uploading to YouTube based on which channel it is. And that'll include all the data around what I want to go like to YouTube. So, that's pretty cool and I'm excited about it because it doesn't like I don't really need to use any tools like I can just use the stuff in here. Um, so adding in ability to manage my YouTube account, manage my YouTube videos, pretty helpful. And then yeah, it's just listening. So then the next one like that just deals with kind of the admin work around it. But the next thing that I'm going to build is actually the uh the thumbnail generation of it. So like that's one that's missing. How do you can you how could you get really good thumbnails created? And then after that it comes into like new idea generation. So having it to be able to pass in the people that make content that I like and go read it, see what's doing well, see what's not doing well, see what's out there, see what's trending, see what people are saying in the comments, store that maybe somewhere, but I doubt it. Um, and then you kind of have a flow where, okay, well, here's, you know, this guy AI Jason, someone mentioned him in one of my videos. I looked at his stuff, and I had seen him a long time ago. he makes just cursor videos and they're great. But if I see one that he's doing, I'll bet that there's something that I know that he doesn't. So then I can say, okay, cool. This is one that's working, so it's validated. So if I don't have a video idea, then that's going to go into a sheet. And then I can decide, do I want to move forward with this? And if I want to move forward with this, then I can tag it. I say, this is a coding video because I make coding and I make marketing automation videos or AI workflow videos. So then it's going to go from the idea down to the idea plus up. So if it is based on someone else's thing, I mean that's what a lot of YouTuber content creators do is like this like re research, right? So I would have it go and actually scrape through the comments and kind of the beauty of again doing it on GCP is I can do it 10,000 times a day and I don't have to worry about what it costs because it's like 0.000000. like I could do it 10,000 times a day for really any operation and it's just because you don't have issues with pricing. You might as well just blast it. Like it's crazy. Uh you can also do sentiment analysis. So this is something I'm doing in our school which will soon be on iviwithai.com. but like mining the information and the sentiment out of what people are saying and then getting the questions, open loops, suggestions, pulling that in. And then the third one is the perplexity plus up. This actually probably wouldn't be perplexity uh but some some search thing that like takes the sentiment analysis, takes the areas that the comments have, takes the YouTube transcript, and then pluses it up. I think that's pretty cool. But where the real sauce will come in, I think, is actually taking it and then doing like the deep research. So again, I put AI Jason's thing in here because I don't know anyone else that makes cursor content. It's kind of crazy, but then it's stage. So it's like cool, it's plused up and the sentiment, let's say they're sad or happy, whatever. And then you have this block of the additional context. Then I want to take all that and say, okay, well, what are use cases for marketers? What are use cases for coders? Well, here's all this stuff. Find three practical high value use cases based on, you know, all this context. So, I think this is going to be pretty cool. And I didn't really even intend on doing it in Google Cloud Platform, but then I started doing the research and being like, what does it unlock if I do it this way? If I spend a couple hours figuring out how to do it, well, it unlocks pricing power. It unlocks me learning new stuff because I frankly don't like clicking around in the drag and drop stuff. And um yeah, it just aligns with what I like doing. So, I'm getting better at at using the tools that I like. So, pretty cool. And of course, there's just like infinite stuff in here. I feel like a lot of people downplay it where GCP is kind of kind of badass. Like if I go in and let's say I want to make a I don't know let's let's take a prompt. Let's make a thin wrapper. So if I went into Vertex AI and I deleted a bunch of projects today so I probably have to reenable this but you have the bottle garden. I think you get $300 if you sign up with a new account. It's pretty cool. And more accurate weather forecasting. What? But you have all the models in here. Everyone knows that 25 Pro is the goat. But I could go in and if I wanted to make a new one, I just hit create prompt. And yes, because this is a brand new account, need to enable. You can see it's now spinning up this resource. But then I can just change the model over here. I could see flash preview. I can see 25 prox 325. So I would probably click on this one. And then I can select if I want a structured output, which is really cool. So, it's like if I'm entering in something and I don't know, let's say I want to take the comments, scrape the comments, hit an endpoint that's hosting three uh 25, return it as this. Return it as like was it good, bad, 1 through 10, how how did the people feel? What was an outlier comment? I can put all this stuff in here. That's what a structured output is. And then that way when I spin this up as an instance or something that I can hit from one of my serverless functions, which is what Cloud Run is, Cloud Run's basically versel if you think about it. And so anyways, I could just put a prompt in here and I would say, hey, like we're doing sentiment analysis on this. Here's a bunch of comments. Go figure it out. I'd say, "Hey, you're an expert at sentiment analysis and I'm going to pass you a JSON blob of all these things and you're going to go and you're gonna figure out what's going on in there." And honestly, there's probably an example of this. I would even have to write this, but I'm just thinking there's probably better algorithms, but that was the use case because I wanted to tie it to something I previously spoke about in the video. But it's just bananas because then you can do this and then if you wanted to like say you had just this like thing this like one prompt well you just hit this and you could deploy as an app like that's kind of nuts and I've said that before but I just haven't spent that much time in the like other areas of GCP. You know a lot of like the DevOps guys are like cool dude just figured this out. Yeah, I had never really used it. Kind of kind of crazy imaging. I can make a song. So, yeah, I'm I'm excited because I want to just kind of map out like that's nuts. I want to kind of map out all the different things that are going to happen and it just has to do with where the content's going to go. And then because like I just want it all one spot because when it comes to the content stuff that's one thing then it's like okay I want to build these agentic workflows that can go and answer questions that I know answers to with my YouTube account on other people's accounts. Oh, I want to go and create the YouTube transcript chapter markers on someone's account using my account. If I can do that, you know, if I have 10,000 times a day that I can max out on for the YouTube API, that's a lot of chances even if I split that across like a lot of different things. There's just a lot. And again, you wouldn't be able to do that um had you not done it in here. Whoa, now I'm talking. What's going on? I need to get out of here. Prompt gallery. Now I'm just like a salesperson for Google, but it's just it's hard to ignore. It's really hard to ignore. And after a couple hours just getting used to the other features, just like little things like I don't know, we're all used to END files, but now you need to have like a separate secret manager for your like refresh tokens and some little things like that that I just didn't know how they work. But very very cool. And what I what I'm looking to learn in here actually is how you map things out. So you always see like DevOps guys and they have like the icons of the different things and I think that's actually where I started with this. If I go to Grock, I started with Grock and then I moved into using cursor but it was like if I if I went back in time to Yeah. five hours ago and I scrolled all the way to the top. Let's just take a look. Yeah, I'm building this out. How would I do it? How would I get a Google Cloud Storage bucket mounted to my Finder window on Mac? Right? And so I go and I do that. I get this Caskfuse thing. I use Go to install it. Now I have that. Then after that, I need to set something up so that it's tied in so whenever my computer restarts, it automatically gets in there. Then I need to figure out how to give my CLI full access to the GCP cloud without it ever having to do anything else. And then that turns into, okay, cool. I got that done. It's like how do I make this thing better? How do I make it so that it can do one of automations for me? So then that's what this one was I think. Yeah, this is when I actually got into the Python and using the artifacts and it's very very cool where I just I think it's going to be crazy like yeah endless endless possibilities. That's enough about that. What uh what happened beyond that? I mean, I've just been in a hole working on this to be totally honest. Let's see what people are asking on the YouTube channel. We'll do that real quick if you are still watching. Um, looks like my video is still not live. Shucks. I made one last night, but it seemed like I don't know. I was filming on potato like I just did a video kind of half-hazardly, and that's okay for the daily channel, but yeah. Let's see. What's the OCR thing that you use? Would appreciate access to the treasure chest. Yeah, I say I gave all these things away and then I'm hoping to launch I Vibe with AI tomorrow. I've had my hands full. Like the community is popping off. I really dislike the offer. I I think the offer is awesome, but I dislike the platform. So, I rolled my own. That'll be really helpful for everybody because it'll contextualize everyone's stuff that they want to learn. Uh, a lot of the people in here have actually joined this community, which is cool. Um, great video. Cool. What about the daily channel? We're about to hit 700 subs. That's pretty cool. 300 views. I didn't even talk about anything in there. What do they say? Uh, community, keep them coming. I think what you're looking for is Wap. No, regarding Grock adding workspaces, I'm very tired of manually version controlling the various files I have in workspaces blah blah blah blah blah. Uh I'm of half of mine to just do everything in cursor because I can switch models without losing context there. Or maybe just start using one of the rappers. Yeah, I don't use coding. Yeah, I don't code in any of the LMS like in their browser. Like I uh today with Grock was really the first time and a lot of it had to do with me not even meaning to like you saw I was just like hey I want to start with this one little thing of just mounting this and then it was like whoa guess what if I just kept going a little further I could just unlock this whole thing but then it got to a breaking point where I was running around doing one debug bug thing and at that point it's like get me out of here. Like artifacts are not meant for anything beyond exploration, prototyping, but it's great at that, right? Like it it it's exploration at the end of the day. Like I wouldn't have done any of this stuff if I didn't start with this flow. So yeah, that's uh I just immediately am trying to get like out of there. Like even if I open this up, right, it's like, okay, I don't ever want to click this thing. I did hear the founder of Verscell, who's great dude, but he's like, yeah, like one of these companies had 127 versions in their VZ. And I'm like, that sucks. That doesn't sound fun. That sounds I mean, good for them, I guess. Don't listen to the haters. Yeah, never listen to the haters, dog. Looking forward to both your channels run quickly. Thank you, sir. I get out of browser and into cursor ASAP. Yep. About this in vid going up in a sec. Yeah. What else? I think everyone just wants stuff and I feel bad because I ended up building a product to host all of it. But yeah, that's it. That's that's the session for today. That's all I got. I appreciate you guys watching. And honestly, these videos are going to get a lot better as soon as I have these systems in place. like even like the pulling in of the news because when I was doing these videos at first I didn't have the second channel and I didn't have the community starting to pop and so doing that on top of the builds just been a lot. So I have kind of back burnered this knowing that I'm going to automate a lot of it. When it comes to the admin work around it and then the research as well where it's like cool on the news side I typically have like five or six resources. They're literally just Twitter lists and I just go through Twitter and it's like boom boom boom boom boom. So if I have a lot of news it means I was on Twitter way too long. That's basically it. So, I'm trying to make something that can pull all that out. Like, actually, before I end it, I made a What did I make? I tried making something in here. Uh, this is our Discord, by the way. What's it called? Big AI news. Yeah, I tried making something that would automate this and it's just like, okay, like, oh, it's just not it's not sick. So, then I just I'm thinking like it's just got to be cool and I don't want it to be totally automated. So, if I can feed it a feed of the stuff that I'm looking at already and just say here's the feeds that I check give me like the best five that probably work. But for now, I'm going to keep building on the automation side so the videos get better. I'm gonna build on I vibe with AI because it's really close and it's going to provide better member experience. It's going to provide all the freebies with free updates. So, that's kind of the why. It's like cool, I can get some emails out of it, but also like it's fair trade because they get a free update because the things change all the time and they don't even have to get the email. So, yeah, I'm excited about it. I'm excited to onboard all the members onto the new thing when I do complete it because we can take all the I don't know when you're in a closed ecosystem you can't like leverage all this AI stuff. So, it' be incredibly ironic if you had an AI community trying to provide AI educational material. You're not using AI. Like, what? You know, low key, I asked the team at at the I don't want to bad about them. Forget it. It's all good. I'm just going to keep building stuff. All right. See you.