Hacks

went to another hackathon. this one was focused on voice agents. it was at the south park, san francisco and around 100 ish people showed up. lots of teams building different voice ai applications. some were interesting, some were obvious, most were somewhere in between. i teamed with a software engineer, works at meta in the bay area, we worked on alarm chat. something i've been thinking about on and off for a while. thought this was a good chance to actually build it. the idea is simple. your alarm talks to you as you're waking up. not just beeping. actually having a conversation with you. it could tell you about your schedule, the news, any research topics you've been diving into. it could read your diary entries, your notes, talk about what's happening on your social media. whatever you need to hear in the morning. the whole point is, you get up better when you're starting a conversation. it's kinda how my dad used to wake me up when i was a kid. gentle conversation instead of jarring noise. and you become more aware of what's going on around you right from the moment you wake up. i've been thinking about voice agents for a while now, and after seeing what people built, i think they fall into two categories. one, human-triggered. you press a button, activate the ai, it does its thing. pretty straightforward. two, automation-triggered. the ai activates itself based on some condition. time-based, like calling you every morning. or event-based, like triggering a call after someone fills out a form. saw a few interesting projects. one team remade podcasts to take in listener questions. basically, the ai recreates the podcast episode, becomes the host, and pauses whenever the listener has a question. answers it, then continues. makes podcasts interactive instead of just passive listening. another one was called the 5 minute chat. an ai calls you twice a day. morning and night. talks to you about your day for 5 to 10 minutes. it remembers previous conversations, builds context over time. not just a one-off thing. but here's what i noticed. pretty much every voice agent needs some context or task or question to work with. it needs a reason to activate. patience is key when dealing with these. you have to wait for the ai to process, respond, work through the conversation. about the judging. i'm noticing a pattern at these hackathons. the judges aren't looking for professional-grade, complex systems. it's surprisingly lean. simple, often obvious ideas tend to win. the ones that just work and are easy to understand. so when you try building something very complex, you usually can't finish it in time. better to pick something simple that you already understand well.

had a good day at the hackathon. around 200 people showed up. lovely food, lovely people. worked on something called "mutual connections" for this one. the sponsors were weaviate, neo4j, vapi, coderabbit, friendliAI, and sovren. we had to integrate them into whatever we built. so here's the idea. an application that analyzes your social media posts. linkedin, x, threads, whatever. it reads through what you share, what you talk about, figures out your interests, your ideas, your personality. it also takes into account what you're actively looking for. like what types of people you want to meet, what kind of conversations you're interested in having. and when you register for an event, it matches you with other attendees. people with shared interests, complementary skills, aligned goals. it suggests who you should meet. highlights mutual connections, interesting overlaps. makes intros easier, more contextual. the whole point is to make event networking less awkward and more useful. you show up already knowing who's worth talking to and why. personalized matching before you even walk through the door. i realized something while they were judging. most hackathon judges aren't looking for some polished, feature complete product. they're looking for three things. does the app work? like the basic version, the minimum viable product. how are you presenting it. and how well did you integrate the sponsor tools. that's pretty much it. also i’ve taken a little note on judging, so once you understand what the judging criteria actually is, you need to position your project around that. make it easily understandable. make it score well on their rubric. most people, myself included, we get an idea, a vague one and then we try to flesh it out. write up some product requirements. start building features. figure out the presentation later. but that's backwards. instead, picture the end first. like actually visualize it perfectly. what are you going to present. what's the presentation going to look like. what functions and features are you going to show. what needs to work for the demo to make sense. then build backward from there. start with the presentation in your head, the exact flow, the exact features you'll demo. then build only what supports that vision. was fun, someone built a product called “amazon john” basically he connected a visual model to amazon account so he can look at something and say “buy me this” and it will actually check it up on amazon, adds to the card, uses his credit card and purchases it. later, i got into a conversation with this psychologist who was sitting in the audience. he's around 80, from new york, now chilling in the bay area. we went deep on human psychology, the basic needs, and all that. i've been thinking a lot about how AI is letting us rebuild everything. all the software, all the systems we've built. and if we're rebuilding from scratch, shouldn't we go back to first principles and think in terms of what actually makes humans tick and what are the root human needs, intentions, the natural language patterns we use. that's what i asked him about. his answer was blunt. he said humans are basically living for survival. that's it. they only care about their own survival. they don't care about the rest of the people. harsh. but it made me think. if survival is the core drive, then everything else we assume about human motivation, all those other layers, they're just built on top of that base instinct.

the hackathon was hosted in the rubpod’s san francisco office, it’s a startup providing gpu services on demand. the hackathon was in collaboration with runpod and comfiui’s team. when i got to the place, it basically looks like a penthouse and got super pumped about how folks in sf hosts events in a penthouse but turned out it was runpod’s office, it’s a startup providing gpu services on demand. the hackathon was in collaboration with runpod and comfiui. so i teamed up with one of my acquaintance from different hackathon. (sf feels so tiny that you run into people that you already met at all these other events) and so we had actually around 90 minutes to use these two tools and comeup with a picture generated the hackathon was hosted as a model called seed dream which was apparently doing better quality image generations than gogle’s nano banana and so the hakathon was hosted by runpod in collaboration comfi ui. run pod was a cloud gpu provider, and comfi ui was an on device application by which we can mix a combination of pictures, text or a video in order to generate a single output. in order to participate in the hackathon, it’s mandatory to use the model seed dream which obviously we had to run it ourselves and so we had runpod and confi ui give us free credits ~ $50 each so that we can run the model seed dream either in the cloud or on device, generate images or videos and participate i the hackathon. this is how usually hackathons be also. in order to push people to try their products they pool up some money, host an event, buy some food & drinks and give the prizes. it’s pretty much a cardinal flow. we wasted a lot of time. thinking what ideas to depict or things like that. and funny thing, we had about 90 minutes to make the image/video and we spent like 60-70 minutes on making some stupid pictures, and talking about ideas, and then we eventually end up making something, which i did not feel like even demoing. but still, it was fun, i liked it. good food, some good people, and damn, i met couple folks dropping out mid way of their phd to go try figure out building things with AI. maybe building things is far fetched better than getting a phd. especially at this particular time. maybe.

I participated in a hackathon hosted in Seattle, focused on AI agents. The whole thing was pretty open-ended, we could build whatever we wanted within 6 hours. It was kinda intense but super fun!What We BuiltI teamed up with Goldwayne and Bharat to create this thing we called "Shadow Agent" basically a productivity assistant for macOS that watches your screen and takes actions to keep you focused and productive. We had big plans for it like notifications blocker, focus timer, reminders, tab manager, app blocker, and even an insights dashboard. However, time flew quick(we spent lot of time debugging) but still, with only just 6 hours in hand, we managed to ship two core features: the notifications blocker (turns on Do Not Disturb mode) and a focus timer with reminders (gives you a nudge when you get distracted). The Cool Stuff I LearnedThe hackathon was a good experience because I picked up some new concepts and different ways to work with AI. I usually have this workflow with Cursor (an AI coding tool) where I discuss the project, come up with phases, and ask for explanations before and after implementation of each phase, Then I'd copy all that to my Notion page for reference.But then [**Goldwayne L.**](https://www.linkedin.com/in/goldwayneleh/) taught me something that made me think "damn, why didn't I think of that?" He suggested asking the AI to create a [**project.md**](http://project.md/) and [**output.md**](http://output.md/) file. Basically, the AI writes down all the phases in [**project.md**](http://project.md/), and when implementing stuff, it references this file. And [**output.md**](http://output.md/) is just where it updates what it's done so you can check it anytime. This is actually smart. thanks Wayne! This project is actually something I like to work on in my free time, so I'm planning to keep working on it and eventually ship a fully developed macOS app with all those features we initially planned. I'm pretty excited about where this could go!Oh, and btw, I drove from Vancouver to Seattle to attend this hackathon. It was a fun good two hour drive, totally worth it for the experience!Overall, I'm pretty stoked about what we managed to build in such a short time, and I learned some cool new tricks working with AI. Definitely a weekend well spent!

as i was surfing through luma, eventsbrite and similar applications, i found this interesting hackathon. it’s a marketing hackathon and i had soo much in my head. one, it’s happening in san francisco. and two, it’s marketing ‘hackathon’ so what’s rolling in my mind was the usage of AI in marketing and so the Hackathon. it’s pretty straight forward. sounds interesting. i’m in. i thought all the folks over there would bring out different techniques, tools, crazy ideas and all sorts of that. so i’m jacked up to interact with all these folks and learn what ever i could. well, it was not exactly what i thought it would be. but it was fun and ofc i had a good time. the whole vibe & experience pushed my thought of the AI bubble a little further. i’ll talk about two things here. one, the whole concept of this and these kinds of hackathons. two, the push of the thought of AI bubble we are living in. one. the hackathon concept was simple. bunch of non profit organizations with in coordination with american marketing association and golden gate university hosted an event called “marketing hackathon” in order for the non profits to get back on to their feet, get some traction online & offline. they basically want people to visit their places. about 35-40 people showed up, we were divided into 6 groups, each got a non profit to work with and were told what they were facing with. we had about 3 hours to come up with an idea, a strategy, design anything that we could, make a presentation and present it. our team worked with SS Jeremiah O'Brien, it was one of the ships from the times of world war II which is still operational (only for tourist purposes twice a year) other wise it sits on the pier 39 in san francisco for people to visit for a price of $45. they are not getting as much of the traffic as of the last year. we came up with some cool ideas. like we noticed their socials were not operational, gave a strategy, i redesigned their website (used lovable) and helped here and there technically and gave a presentation. went well. but it kept me thinking of this is actually a very smart way of getting a set of people to work on your unique case and get to a strategy to implement. pretty much like a mini consultancy for almost at no cost. well yeah, they can ask chatgpt as well but the people who are working over there are in their 50’s and 60’s. it would be a back and forth process with chatgpt and ai usually overwhelms by dumping too much information and too many things to do. and almost all of us already visited the place already before, we know the surroundings and could try giving a decent strategy understanding their budget constrains. two. i strongly felt that the tech bros are living in the ai bubble. like if you are on x (twitter) or watching closely of the AI labs & some interesting AI startups doing, you’d see new models, new features, new techniques, new ways of doing things, and would feel like there is just not enough time or a capacity to know or learn about all these and would also feel like you’re lagging behind. fomo hunting you. but, if you step aside, get into different fields (even in the silicon valley) the adaptation curve is slow. very slow. like not lotta folks out of tech hardly uses ai in their daily lives for study, work or in general. felt strange to see that in san francisco. but it’s normal. the tech bros are in the bubble for sure. the transformation of the world being ai first will take time. and btw, they also gave me a certificate. cute. 

I made an application solo at a hackathon, of course I have used a lot of AI. This was hosted by Red thread Club, partnered with ALly Global, Lovable This was an interesting one, the case was presented around 30 minutes before the hackathon begun, and we had 4 hours to finish it. Interestingly, I spent more than two hours just planning. Like, firstly I took a good chunk of time understanding the aspects of the case presented, I did talk to Claude, Gemini, Grok and Chatgpt about the different ideas (Gemini 2.5 pro gave me the best idea for me to iterate) then, i went on to choosing the tech stack, the data base, the way data flows, the areas of LLM integration, the prompt on which the LLM responds to, the UI, the presentation and demoing. I divided the implementation into phases and watched cursor implement each phase and asking it a million questions as it is making the application.(i used Gemini 2.5 pro and Claude opus 4 as the models inside cursor) For presentation, I just asked cursor to give me the summary of all the features that it has implemented in separate paragraphs, copied them and gave it to Gamma. and it gave me 9 slides in under a minute, deleted a few slides for simplicity and called it a day.