DEV Community: Seb

This AI Is a Better Geography Teacher Than Most Humans.

Seb — Sat, 13 Sep 2025 15:45:51 +0000

This is a submission for the Google AI Studio Multimodal Challenge

What I Built

I built GeoGenius, an AI-powered interactive learning companion designed to make learning geography dynamic, engaging, and fun. The problem it solves is simple: traditional learning from static textbooks can be dry and fails to capture the dynamic nature of our planet. GeoGenius transforms any geography topic—from a simple text prompt or even a photo of a textbook page—into a personalized, multi-faceted learning module in seconds.

The user flow is straightforward:

A user enters a topic like "Causes of Urbanization" or uploads their study notes.
They click "Generate Lesson."
GeoGenius uses AI to create a rich, structured lesson plan that includes:
Key Concepts: A clear, concise breakdown of the most important ideas.
Simple Analogies: Complex topics are related to everyday life to make them easier to understand.
Visual Moments: Each analogy is paired with a unique, AI-generated image to visually reinforce the concept.
Interactive Animations: For dynamic processes like the water cycle or tectonic plate movement, a simple, hands-on animation is provided.
Test Your Knowledge: A short, multiple-choice quiz helps the user check their understanding and solidify what they've learned.

It's not just a Q&A bot; it's a creative partner that builds a personalized and comprehensive lesson from scratch, every time.

Demo

link to the applet: https://geogenius-interactive-geography-lessons-598974168521.us-west1.run.app

For a complete walkthrough, here is a short video that shows the entire process from entering a prompt to interacting with the final generated lesson.

How I Used Google AI Studio

Google AI Studio was the core in the development of GeoGenius, particularly for prototyping and refining the core AI prompts.

My workflow was centered around structured output. I knew I needed a reliable JSON object from the AI to render the different components in my React app.

I used the AI Studio playground to:

Craft the System Prompt: I iterated on the main instruction ("You are a world-class geography teacher...") in AI Studio to find the perfect persona and set of instructions for the model. This ensured the tone and quality of the content were consistently high.
Define and Test the JSON Schema: The "Structured Prompt" feature was a game-changer. I designed my entire lessonPlanSchema directly in AI Studio, defining the types, required fields, and even providing descriptions for each property. This allowed me to test prompts and see how the gemini-2.5-flash model would populate the schema, ensuring the output was always predictable and ready for my frontend to consume.
Prototype Multimodal Inputs: I also used the playground to test how the model would respond when given an image (like a page from a textbook) along with a text prompt, which helped me build the file upload feature with confidence.
After perfecting the prompts and schema in AI Studio, I simply transferred that logic into my application code using the @google/genai SDK. This process saved hours of development time by allowing me to separate prompt engineering from application logic.

Multimodal Features

GeoGenius leverages multimodality to create a richer and more effective learning experience than what text-alone applications can offer.

Multimodal Input (Image + Text → Lesson): The app’s ability to accept an image (like a .png or .jpg of a map or textbook page) alongside a text prompt is its most powerful multimodal feature. A student can literally take a picture of their homework and ask the app to build a lesson around it. This turns static, offline content into a dynamic and interactive digital experience, meeting the user exactly where they are.

Text-to-Structured Data (JSON): At its core, the app transforms an unstructured text request into a highly structured JSON object. This is a crucial modality that allows for a reliable and sophisticated UI. Instead of getting a single block of markdown, the AI provides a predictable data structure that the React frontend can map to distinct, beautifully styled components for concepts, analogies, quizzes, and more.

Text-to-Image Generation: To make abstract concepts more tangible, the AI first generates a creative visualPrompt for each analogy in the lesson plan. My application then takes this text prompt and uses the imagen-4.0-generate-001 model to generate a custom illustration. This on-the-fly image generation ensures that every lesson has unique, contextually relevant visuals, which are far more engaging and effective for learning than generic stock photos.

Can an Algorithm Dream? I Built an App to Find Out.

Seb — Tue, 09 Sep 2025 04:34:53 +0000

This is a submission for the Google AI Studio Multimodal Challenge

What I Built

I built the Progressive Story Maker, an interactive web application that transforms storytelling into a collaborative, choice-driven experience with an AI.

The app solves the problem of "writer's block" and creative inertia by turning narrative creation into an engaging game. It begins by generating the first sentence of a story in a user-selected genre (Medieval Fantasy, Modern Mystery, or Kiddish Adventure). Within this sentence, key words are highlighted. When the user clicks a word, it becomes the creative prompt for the Gemini API, which then generates the next paragraph of the story. This new paragraph has its own clickable keywords, allowing the user to continuously guide the narrative down unique, branching paths. The result is an endless story machine that empowers users to co-create completely original tales simply by making a series of simple, intuitive choices.

Demo

here is the link: https://progressive-story-maker-598974168521.us-west1.run.app

How I Used Google AI Studio

While the final application interacts directly with the Gemini API via its SDK, Google AI Studio was an indispensable tool during the development and prototyping phases.

Prompt Engineering & Refinement: I used the AI Studio playground extensively to design and test the prompts that power the application. It provided a rapid feedback loop for crafting instructions that could reliably generate compelling story segments and, most importantly, extract exact, verbatim keywords from the generated text. This was crucial for ensuring the frontend could always find and highlight the interactive words.

Structured Output (JSON Mode): AI Studio was instrumental in defining and validating the responseSchema for the Gemini API calls. By experimenting in the studio, I finalized a robust JSON structure ({ "paragraph": "...", "keywords": [...] }). This use of structured output makes the application incredibly resilient by guaranteeing that the data received from the API is always predictable and correctly formatted, eliminating the need for fragile string parsing and significantly reducing potential runtime errors.

Model Selection: I used AI Studio to evaluate different models, ultimately selecting gemini-2.5-flash for its optimal balance of speed, creativity, and cost-effectiveness, which is essential for a real-time, interactive user experience like this one.

Multimodal Features

The current version of the Progressive Story Maker focuses on perfecting a masterful unimodal (text-to-text) experience to ensure the core narrative mechanic is seamless and engaging.

However, the application was architected as a strong foundation for future multimodal expansion.

Image-Driven Prompts (Image-to-Text): To introduce multimodal input, users can upload an image instead of clicking a keyword. The contents of the prompt sent to Gemini would then include both the image and a text instruction like, "Continue the story based on this image." This allows users to introduce completely new visual concepts into the narrative, giving them an even more powerful way to guide the AI's creativity and making the experience truly multimodal.

The Ultimate Life Hack for Turning a Boring Walk into a Real-Life Video Game Hometown

Seb — Tue, 09 Sep 2025 03:36:44 +0000

This is a submission for the Google AI Studio Multimodal Challenge

What I Built

I used Google AI Studio to make City Quest AI is an interactive, AI-powered scavenger hunt application that transforms any city or neighborhood into a dynamic adventure. The app solves the problem of generic, pre-planned city tours by creating unique, on-the-fly quests based on user input. Users simply enter a location (e.g., "Downtown Seattle") and an optional theme (e.g., "Historic Landmarks" or "Coffee Shops"), and the app generates a multi-stop scavenger hunt complete with clever clues, fun challenges, and interesting facts for each location.
It's designed to encourage local exploration, create fun and competitive experiences for groups of friends, and make discovering a new place more engaging and memorable.

Demo

a link to my deployed applet: https://city-quest-ai-598974168521.us-west1.run.app

Setup Screen: The user is greeted with a sleek, modern UI. They enter their desired city and an optional theme for their adventure.

Loading Screen: After submitting, a loading animation appears while the Gemini API crafts a custom scavenger hunt in the background.
Game Screen: The main interface displays the first clue. Once solved, the user can reveal the location, which appears on an embedded map. They then complete a fun, interactive challenge (like taking a photo) and learn a fun fact about the place. The screen also features a live scoreboard, adding a competitive element to the game.

Completion: After finishing all the stops, the user sees a congratulations screen celebrating their victory and encouraging them to play again.

How I Used Google AI Studio

I leveraged the Gemini 2.5 Flash model through the @google/genai SDK to serve as the creative "game designer" for the app. The core of the integration lies in using structured output capabilities.

In the geminiService.ts file, I defined a strict JSON schema (responseSchema) that outlines the exact data structure needed for each scavenger hunt stop: clue, locationName, challenge, and funFact.

By providing this schema in the API call along with a detailed prompt, I instruct the Gemini model to return its creative output as a perfectly formatted JSON array. This is powerful because it eliminates the need for unreliable string parsing and ensures that the data from the AI can be directly and safely loaded into the application's state, making the integration seamless and robust.

Multimodal Features

The primary AI feature is Text Generation with Structured Output. The application takes simple text input from the user (location and theme) and uses the Gemini model to generate a complex, structured JSON object that defines the entire game. This goes beyond simple text generation by forcing the AI to think and create within a specific, developer-defined framework.

While the AI generation is text-based, the application creates a multimodal user experience. The AI-generated challenge text often prompts the user to interact with the physical world using their device's camera (e.g., "Take a team photo in front of the fountain"). The user then uses the built-in photo upload component, which accesses the device's camera, to provide visual proof of completion.
This creates a fun loop:

Text-to-Game: AI generates a text-based challenge.
Real-world to Image: The user interacts with the real world and captures it as an image.
Game Progression: The image upload confirms the challenge is complete and unlocks the next part of the adventure.

Give it a try!

Seb — Tue, 09 Sep 2025 03:27:15 +0000

Google AI Challenge Submission

Seb

Sep 6 '25

Don't Wait for an Accident. This AI Tool Spots Hazards Before Your Baby Does.

#devchallenge #googleaichallenge #ai #gemini

Comments 4

3 min read

Give it a try

Seb — Tue, 09 Sep 2025 03:25:06 +0000

Google AI Challenge Submission

Seb

Sep 7 '25

This AI Guesses Your Drawings Faster Than Your Friends Can.

#devchallenge #googleaichallenge #ai #gemini

Comments 5

2 min read

This AI Guesses Your Drawings Faster Than Your Friends Can.

Seb — Sun, 07 Sep 2025 16:21:06 +0000

This is a submission for the Google AI Studio Multimodal Challenge

What I Built

I built an interactive web application that brings the classic drawing and guessing game into the digital age with a modern twist. The app challenges a player to draw a word provided by the game, while a sophisticated generative AI attempts to guess the drawing in near real-time.
This creates a unique and engaging solo-player experience where the user's artistic skills are pitted against the AI's image recognition capabilities. It solves the problem of needing multiple players for a game of Pictionary and provides a fun, interactive way to experience the power of multimodal AI.

Demo

link: https://ai-pictionary-598974168521.us-west1.run.app

How I Used Google AI Studio

I leveraged the Gemini API, accessible through the @google/genai SDK, to power the core guessing mechanic of the game. Specifically, I used the gemini-2.5-flash model for its speed and powerful multimodal capabilities.

The implementation involves capturing the user's drawing from the HTML canvas as a PNG image, converting it to a base64 string, and sending it to the Gemini model. This image is sent alongside a carefully crafted text prompt: "What is this a drawing of? Look at the image carefully and provide your best guess in a single word." The model then processes this combined visual and textual input to return its guess as a single word of text. This demonstrates a powerful image-to-text, or visual understanding, use case.

Multimodal Features

The central multimodal feature of this application is visual reasoning and description. The app seamlessly integrates two distinct modalities:

Image Input: The user's free-form drawing on the canvas serves as the primary visual input.

Text Output: The Gemini model analyzes this visual information and generates a textual guess.

Don't Wait for an Accident. This AI Tool Spots Hazards Before Your Baby Does.

Seb — Sat, 06 Sep 2025 05:03:08 +0000

This is a submission for the Google AI Studio Multimodal Challenge

What I Built

BabySafe AI is a user-friendly web applet designed to give parents and caregivers peace of mind. It solves the often overwhelming problem of childproofing a home for a mobile baby (6-18 months) by acting as an AI-powered "second set of eyes."
Users simply upload a photo of a room, and the applet leverages the Google Gemini model's vision capabilities to analyze the image, identify common household safety hazards, and present them in a clear, actionable list. The experience is designed to be simple, fast, and reassuring, turning a potentially stressful task into a straightforward safety check.

Demo

link: https://babysafe-ai-598974168521.us-west1.run.app

Here's a walkthrough of the BabySafe AI user experience:
Upload: The user is greeted with a clean, simple interface where they can drag-and-drop or click to upload a photo of a room.
Preview & Analyze: Once an image is selected, a preview is shown. The user then clicks the "Analyze Safety" button to begin the AI analysis.
Analysis in Progress: A clear loading indicator lets the user know the AI is working.
Hazard Report: If hazards are found, the results are displayed in a two-panel layout. The original image is on the left, and a detailed, scrollable list of identified hazards is on the right. Each item in the list includes the hazard's name, the specific risk it poses, and a clear description of its location.

How I Used Google AI Studio

I used the gemini-2.5-flash model from the Google AI platform to power the core functionality of BabySafe AI. My implementation focused on two key multimodal capabilities:
Vision Understanding: The primary function relies on the model's ability to interpret and understand the contents of an image. It goes beyond simple object detection to recognize items in the context of child safety.

Structured JSON Output: I used Gemini to write up a detailed system prompt and provided a strict responseSchema. This instructs the AI to act as a "Baby Safety Expert" and forces it to return its findings in a predictable JSON format. This was crucial for reliably parsing the AI's response and rendering the hazard list in the user interface. By defining the expected data structure, I turned a powerful, general-purpose vision model into a specialized and reliable analysis tool.

Multimodal Features

The core multimodal feature of this applet is Image-to-Structured-Text Analysis. It takes a visual, unstructured input (a user's photo) and transforms it into structured, actionable text data (a JSON object listing hazards).

This enhances the user experience in several key ways:
Intuitive Input: Users don't need to describe their room with words. They can simply show the AI the environment with a photo, which is a far more natural and efficient way to communicate complex visual information.

Contextual Hazard Identification: The AI doesn't just list objects; it identifies them as hazards. It understands that an electrical outlet low on a wall is a risk to a crawling baby, or that a dangling cord is an entanglement threat. This contextual understanding is only possible through analyzing the visual data.

Spatially-Aware Descriptions: The model generates a location_description for each hazard (e.g., "On the floor in the bottom-left corner"). This spatial awareness, derived directly from the image, makes the feedback immediately actionable. The user knows exactly where to look and what to fix, transforming a generic warning into a specific, helpful instruction.

A social media app, powered by BrightData & n8n

Seb — Tue, 26 Aug 2025 10:44:11 +0000

Ok this project is built to eventually be submitted to the Real-Time AI Agent Challenge.

This web application pulls news headlines, it is given to an LLM that is tweaked to exhibit different view points. After this it gets sent as a post on the app for everyone to see.

No one can make a post, but they can comment and show if they agree with the opinion of the model.

Currently this app is like 80 percent there. i just need to setup authentication and the design.

I really want to see this app atleast get 100 people on it. So please let others know.

AI Helped Me Build a Redis-Powered IoT Anomaly Detector. In 1 Day.

Seb — Sat, 02 Aug 2025 01:42:50 +0000

This is a submission for the Redis AI Challenge: Real-Time AI Innovators.

What I Built

While vide-coding, I made a Redis-Powered IoT Anomaly Detector in one day." It demonstrates a real-time data processing pipeline designed for an Internet of Things (IoT) sensor, using Redis Streams and Redis Time Series. The system ingests continuous sensor data streams, detects anomalies through a threshold-based logic, stores data efficiently for historical analysis, and raises alerts for critical events in real time.

Demo

https://github.com/SebastianKibiriti/redis-iot-anomaly-detector-mvp-withGemini

How I Used Redis 8

It uses Redis Streams for real-time data ingestion and inter-service messaging, and Redis Time Series for optimized, long-term storage and analysis. Two key Redis features that are central to Redis 8's capabilities for IoT and time-series applications.

What are Your Daily Driver Apps??

Seb — Thu, 24 Jul 2025 16:00:04 +0000

So Gemini has a feature to use the search history of the user for context. I never thought how I could use it, until now. This post has given me a reason to try out that feature. Considering the last 30 days, I've been pretty using the same apps/software for developing applications. Despite the advances that the competitors make, the tools I use are pretty much comparable to the competition and I don't notice a performance difference. I just use them since I am used to them.

Tools I used in the last 30 days.

Large Language Models/AI Assistants:

Gemini
ChatGPT

AI Image Generators:

Ideogram
Higgsfield

AI Terminal:

Gemini CLI

AI Video Generators:

Higgsfield

AI Development Platform:

Bolt.new
VS Code
Google AI Studio

Making It Up As I Go.

Seb — Fri, 11 Jul 2025 08:46:27 +0000

This is a submission for the World's Largest Hackathon Writing Challenge: Building with Bolt.

My plan is to update this essay for the duration of this challenge. Essentially to write how I felt, what i remember doing during the WLH, rather than submitting a full essay in one go.

The Start:

Wrote on 11.07.25

A HACHATHON that gives you the tools to embrace your creativity really is a big time opportunity to show the world what you can do. So when the builder's pack was being distributed, I was very excited to as I waited to receive my own pack. I felt like I could make any application I wanted, with a Pro subscription with Bolt.new, Elevenlabs, dev.to, an many others, the world was TRUELY my oyster. The desire to make anything that I could think of was possible.

Then when I finally received my own Builder's Pack. It was go time. I already read through the rules and understood the challenge. Now, I had the toolbox. I got my Pro sub for Bolt.new, then I went on to make my first project. I knew what application I wanted to build, so went I prompted bolt what to build, was very happy to see it build what is requested on the first try. My only problem was that, I didn't know how I wanted it to look - and I'm a designer - so I planned to rather get the application functional first, then build its form afterwards.

The first of four Abandoned Projects: Home Soidier

Wrote on 12.07.25

I was building a sort-of, progression game for kids, with the parents being the role of order. The application would help make chores truly rewarding. I you have ever watched the first season of Everybody Hates Chris, then you would remember an episode, where Chris briefly talks about an odd game his family would play, about cleaning the damn house. In that scene, Chris mentions that he won the garbage removal competition 3-4 years in a row.

Imagine that, for 3-4 years you are the only sibling taking out the garbage because your siblings don't wanna get their hands dirty. An the only reward is a thank you. Sorry, but, NO THANK YOU. At that point, the system feels rigged. Appreciation gets old, a child wants to be rewarded for their efforts. So, my vision for the application was to make the child work through their work, and keep their reward and recognition in their sights. This would make them more diligent in doing their chores without being told, and motivate then to take on more responsibility as they level up, but to use their position properly - like the child with the highest level being next in charge when mom and dad are out, but being demoted if they abuse their role - this making for more motivated children.

So, a website was made for parents to be introduced, and the main activity page that would be used by the child of the parent. I got the authentication setup with Supabase, all the buttons functioning. Now I need to take it a step further, with some cool and innovative, AI. I wanted to have a chatbot that the child would talk to, not chat to, I had an Elevenlabs Creator account, I wanted to make full use of my tools. Apparently, this was not gonna be as easy as I thought. When I prompted Bolt, to allow me to have a LLM that I can speak to and have it talk back to me, I bit off more than I could chew. And this is where all hell bro loose.

Bolt told me that I needed Api's for transcribing, for the model to interpret the message, for elevenlabs to use the realistic voices, and that I need to store then as secrets/or function in the backend. So I took a quick detour to fetch the api keys for each service, I prepared to put them in the .env file, then I found out that is a big no no , and I actually need to store them on the backend as it suggested before. Now I have never used any database before, so instead of blindly clicking on buttons, I use the AI chat on Supabase to do me a solid. I gave me step-by-step guidance, but it wasn't totally easy to follow as it referred me to places that their titles changes, so it left me guess working if I was in the right place. But, eventually I got it done. I could use the microphone, send a message and got back an error message.

Everything worked like I wanted but I just could not talk with the AI chatbot, and to add insult to injury, my preview screen started to seize up showing me a forever loading page, flickering white occasionally. Now, I cannot even see any future changes I make.

Wrote on 16.07.25

At this point I was using a Celeron laptop, 2017 model w/ 4GB ram. So I thought that this might be a hardware problem, and to test it out I visited my bro to use his pc. An HP EliteDesk 800 G3. i7 6700, 8GB RAM, AND....an NVME SSD. I was open to a new world. Blazing fast boot and loading speed. I could continue to work on my projects.

Curiosity killed the cat.

Now I was getting deeper to using bolt, so checked out the documentation for best practices. The two I tried out was clearing unused files and resetting the context window. I did this to experience how it was like, after that i deleted the duplicate projects and continued. However, I noticed that my token were taking a hit, like 200k per change. So I decided to try saving my token as I tried before. But this time, I took a different approach. I went to settings, clicked on general, then I saw the button to delete chat all chats. Then I clicked in, and all of my projects was gone.

Honestly when I clicked this damn button I thought it would do the same thing as resetting the context window. So, when I saw my project was gone, that it no longer existed, I panicked hard. I was distraught, "all my work was gone, all that hard work for nothing" - I know lame, I considered the result of all my prompting in natural language as "hard work" - but I realized that I already integrated Github to this project, so I just had to make a new connection to restore what I lost. There was still hope. But considering that I was proficient with Github as old people are with using their phones, that hope was looking grime.

Now this is exactly what I did.

I went on a new page on bolt.
I selected the import form Github button, I selected the repo that I was working on.
waited

Horray, connection has been restored. I was so relieved. So, I went on to ask bolt to make a change to my project, and as I waited a response came back, saying something like request could not be fulfilled, or failed to load files. It was like getting stabbed in a wound that already healed. I clicked on a bunch of files and they has no code at all. Now, I was so discouraged by this fact, that I decided to move on to a now project, and try remaking this project later.

Quick side story. For the second project I wanted to make, I wanted to try it out in ONE SHOT. I would give bolt the would structure of the website, what the purpose of the website was about, how you would navigate, , the colors, the functions, all of it. And I used Gemini to setup the prompt. once I was ready I, I Ctrl+C, then Ctrl+V'd into bolt, hit go and I kid you not it literally made 99.99% of what I wanted. c And this is the project that I was so happy of that I made a submission to the hackathon, for the single prompt challenge. I called in helloworld. DO you think you can guess what the website is about.

The second of four Abandoned Projects: BoxerBill

** Wrote on 22.06.25**

Now I want tackling my second/third serious project. After the abomination that happened with my first project I learned quite a lot about how Bolt functions and its integrations. So, for this project. The idea was to detect when you commit to a subscription and it will simply make a reminder a few days before the due date to for you to cancel. Then when the due date arrives, you can cancel the subscription yourself or use a browsing agent like Runner H to cancel it for you. And that is what I told Bolt. It was able to make this into an extension that I could download a unpack in chrome.

Now as a rookie, I didn't know the first thing about making an extension. Just getting it to load in chrome was a pain, apparently a dist folder needs to be compiled before you can download the project. So when it did finally load, I just wanted the detection feature to work. AND IT DIDN'T!!!! At this point I was ready to crash out, I couldn't even get Ai to understand what Ai should do - I used Gemini to form the input I provided to bolt.

The way that the extension was suppose to work was to register a click on any button that leads to the users subscribing, so like, recognizing the words 'pay now' or 'get plus', and if the user clicks on that button to sets a calendar reminder.

So since this is how I made it operate I thought a simple simulation should help test the feature. So I headed over to Gemini to make such a sim., did the test, and you know the rest.

Free with a fee.

Well at this stage, I made 2 projects. Bolt had Entri who offered free domains, and I wanted to have a website, so I went exploring. There were a handful of names that I thought of, and they all sounded cool. But I committed to one that suited it very well. I proceeded to get my domain which was free for a whole year, when I clicked buy, I went on my banking app to confirm. But, I noticed something strange. I needed to pay a one time admin/or setup fee to get this free domain. So there was a catch. Great. I was not planning of paying for a 'free' domain, so I didn't. I cancelled and looked for another option. The builder's pack had a bunch of goodies that I still had not used of I went through them and I did catch something good.

There was a Dev Community subscription which itself was like a pack of other subscriptions, and there was a lot. So card by card I looked for the solution to my problem. I was able to find 2 services offering free domain, a .fun, and a .tech. I decided to go with the .tech domain because the things I was making in bolt were less 'fun' and more 'tech'. And it was actually free, no hidden costs. I was happy to have to have a clean domain to my liking, now I needed to set up the connection. The process was not all that difficult and I went on to use this domain for all the projects I made with bolt.

After that I moved on to make my next project.

The third of four Abandoned Projects: Following

This was a simple statistical project, and starting off small, I wanted to connect 5 of the most popular social media platforms to show the total number of followers the user had.

I envisioned a big and bold badge that showed the total followers from all platforms, when you get connected. Then some basic metrics and tools. Once again I used Gemini to form the necessary prompt and I pasted it into Bolt. Now the results was mostly what I wanted, but I was so wowed by what I made that I decided ditch what I planned and just work with what I got.

I was pleased to see how relevant the tools that bolt suggested were to the platform, and the demo version worked well. Smooth animations and everything updated accordingly. Now I again faced the challenge of integration data through api to show the follower count and some metrics from posts.

So in my frustration I took a break. For like a week. Then my the end of my break I received an article that talked about a study conducted to assess the impact of using AI. And the article really, really struck me. The group that used ChatGPT as much as they wanted, performed cognitively worse than the group that had zero access to AI. AND worse, that same group did EVEN worse when their access to AI was cut off.

Every Ai tools is programmed to give you the answer first try. So most of us are going to become in some way like that group, especially school students.

So that moment felt like I came up with a big brain idea. "Just make a model that behaves like a mentor". It will guide you to the answer, not just be an common answering machine.

So I got to it.

The last of four Abandoned Projects: Phil

Now around this time it was the weekend, and Bolt gave 20M tokens for a limited time. So I went crazy. Keep in mind the Pro plan gives 10M and I had 3M remaining. I knew these token would go to waste if I was to strict. So in this project is use the 21st.dev components, hard. getting the catchiest hero, coolest animations, cleanest UIs. I was having an absolute field day. Also sorting out bugs was not that hard. The setup for this project was more simple that my other projects. Just 2 pages in total, the main page is probably 90% made with the components from 21st.dev, and the UI for the second was like ChatGPT and Gemini. Very simple. I got it all right and I even was able to chat to the model like I envisioned. I does not give the answer, I guided me to it. And this was my next best project. I was committed to submit this one to the hackathon. I really wanted to use the voice feature with 11Labs's realistic voice but I was happy to make the changes later. All I had to do was make the video and do the submission. I could work on it later if there was time.

So I made the video, I posted it, and I started on the submission process. Now at this point if felt like life just wanted me to pull a fast one on me because during the process of me submitting my network went straight offline. Like a line was severed. I really could not do anything at that point, and there were literal hours till the deadline.

After a while of waiting and hoping, I just put it aside to do whenever the network goes live. But I only had a short window. 2PM comes by, still offline. 4PM comes by, still nothing. 6:30PM comes by, WERE BACK BABY, no it was just an involuntary spike by a network with no hope of coming back soon.

I have a data plan that cuts off at 7PM in the evening until 12 midnight. So I absolutely was in trouble. Because I take a while to fill the pages on the submission form. So there was no room for error it the worst situation played out. But it did! And it was worse!! I woke up late even, so the ABSOLUTE WORST situation was playing out!!! My heart was throbbing, I rushed to check the network status and I watched it for 2 minutes, the connection was stable but so weak. Sooo weak. I went on devpost, and saw there was actually an extension on the headline, so go to it. I regretted not at least preparing the form before have so that I could copy and paste now when I needed it.

Tick tock, tick tock, the remaining time is not a lot. The page too forever to continue, I wondered if it even mattered to continue. But, I continued with the hope that if the connection improved then my chances would be better. But alas, it just wasn't meant to be. Submissions closed and I was unsuccessful.

Reflecting

I had 5 projects made in Bolt, but only one made it to the finish. The second fell just short, and the rest needed more attention, skill.

I remember waiting as I anticipated the things I would make with such a game-changing technology. To make things I could only dream of, like my own Jarvis AI, like Tony. But then when the moment comes, I was stuck wonder what to even make. I started to wonder where are that desire and vision go. Steadily they rolled in, and the only hurdle was the knowledge to use these tools.

The builder pack had a bunch or free goodies and I just barely scratched the surface to really getting to use them. Regrettably, I never even used some of them. A cool debugging tool, Sentry, shows replays of errors that occur to make debugging easier. I didn't get to use it all that much during the hackathon, but it has like a 6 month paid for plan that I activated and I do hope to use it some more.

As of now, little progress has been made on my project as I mostly used by brother machine of a laptop which he now needs like all the time. So I gotta be quick with what I do. Although quick is not that much.

I surely got to get familiar using Github and Git for controlling the version I worked on. At one point I had like 200 unmerged commits, and it was a serious pain merging them and not breaking anything. There were pushes then reverts, then new branches needed to be made.

Another cool tool I missed out on was Pica. Not to be confused with Pika, the AI Video generation tool. I thought it was the same so I didn't check it out until it was tool late :( .

Well that is all for my journey to "vibe coding". If I think of something interesting that happened that I didn't already share I may update this. Or post a follow up/expanded topic.

If you read all this, thanks. Very Interesting to see such a dedicated reader!

all the projects i am working on are at

https://modelinfer.tech/

Never Miss a Local Meetup Again! 🚀

Seb — Sun, 06 Jul 2025 15:49:27 +0000

This is a submission for the Runner H "AI Agent Prompting" Challenge

What I Built

I built a "Local Meetup/Club Discoverer" workflow using Runner H that automates the tedious and time-consuming process of finding new local interest groups and events.

it solves is the challenge of staying informed about new and relevant community activities.

Demo

https://x.com/i/status/1941882936694804961

How I Used Runner H

Here's a breakdown of how Runner H's capabilities were leveraged:

Autonomous Web Navigation: Runner H was instructed to navigate directly to Meetup.com (and potentially other community forum URLs).

Dynamic Form Interaction: I leveraged Runner H's capacity to identify and interact with dynamic web elements. This included inputting search terms into the "interests" field, entering location data

Intelligent Data Extraction: Runner H was tasked with extracting specific data points from the search results, such as group/event titles, descriptions, dates, locations, member counts, and direct URLs.

Structured Output Generation: Finally, Runner H was directed to compile all the extracted, filtered data into a well-organized spreadsheet (e.g., Google Sheet).

Instructions to Replicate the Workflow:

To replicate this workflow, you would provide Runner H with a prompt similar to the one below.

Prompt for Runner H:

"Discover new local meetup groups and events.

Inputs:

Interests: [Comma-separated list of interests, e.g., "hiking, board games, coding meetups"]
Location: [City, State/Province, Country, or Zip/Postal Code, e.g., "Cape Town, Western Cape, South Africa"]
Radius: [Distance and unit, e.g., "25 miles" or "40 km"]

Workflow:

Go to Meetup.com.
For each specified interest, search for groups and events within the given location and radius.
Identify groups or events that appear new or have recent activity (e.g., events posted or updated in the last 30 days, or upcoming events).
Extract the following details for each relevant finding: Group Name / Event Title, Description, Date & Time (for events), Location, Number of Members / Attendees, Link to the Meetup.com page, and any relevant Tags/Categories.
If possible, also search prominent local community forums (e.g., [mention specific forum names if known, otherwise general search]) for similar new group or event announcements and extract relevant details.
Filter out irrelevant or inactive entries.
Compile all collected data into a structured table, with columns for all extracted details, and indicate the source (Meetup.com or specific forum). "

Use Case & Impact

For Individuals Seeking Community:

Newcomers to a City: Helps quickly integrate into a new environment by finding relevant social and professional groups.

Hobbyists & Enthusiasts: Effortlessly discover new clubs or events for niche interests (e.g., "vintage camera repair," "Dungeons & Dragons campaigns") that might be hard to find otherwise.

For Community Organizers & Businesses:

Event Planners/Marketers: Gain insights into the local event landscape, identify popular interests, and discover potential collaboration opportunities.

Local Businesses: Understand community interests to tailor marketing efforts or identify potential target audiences for local promotions.

How it Improves Existing Processes:

Time Efficiency: Dramatically reduces the hours spent on manual searching across multiple platforms, freeing up users for more meaningful activities.

Comprehensive Discovery: Ensures a wider net is cast, increasing the likelihood of finding truly "new" or less-advertised groups and events that might be missed by casual browsing.

Reduced Friction: Lowers the barrier to entry for community engagement by making discovery effortless and presenting information in an immediately actionable format.

Proactive Engagement Instead of reactively searching when a need arises, users can proactively receive updates on new opportunities, fostering continuous community involvement.

Data-Driven Insights: The structured output provides a clear overview, enabling users to quickly assess and prioritize which groups or events to explore further.

This Runner H solution transforms a tedious process into an automated, insightful process, empowering individuals and organizations to connect with their local community more effectively and to better understand and engage with local interests.

Social Love

I post on X. https://x.com/directors_comp/status/1941882936694804961