I joined an AI Hackathon...
Pre-Hackathon and Prompt
I've been looking forward to this Hackathon to be able to meet new people, build an awesome product, and learn from other builders.
The prompt was simple: "AI Runs The Shop". Build an AI business that can autonomously operate. I think it's very straightforward. I got excited, but that excitement quickly turned to anxiety the night before the Hackathon. I started overthinking my weaknesses and lack of knowledge about building AI agents. Up to this point in my career, I've utilized AI tools such as Claude Code, Codex, Gemini, Lovable, etc. However, I've never made an agent. The closest I've done is utilizing OpenClaw to poll every 3 hours to work on a side project together via Discord and Todoist. Let me know if you want to see my setup there. I couldn't sleep until 3 am due to overthinking.
I was able to sleep after asking Claude about different AI architectures such as ReAct (Reasoning + Action) AI architecture and wrote some goals that I want to get out of the Hackathon.
Goals:
- WIN 10K! (Shoot your shot, baby!)
- Lead how I can
- Learn and Build AI Agents
- Have fun and meet new people! Touch grass.
Hackathon, The Team, The Approach
Don't get me wrong. I live 10 mins out of Podium HQ, but the drive felt forever at 8:50 am on Saturday. I was still nervous. That feeling was gone immediately after entering the building. The vibe and environment were very exciting and open. You can feel everyone is just eager to Hack and Build. I also caught up with my friends back in uni, Jae and Aoi.
After the kick off with Eric Rea, Dennis Steele and Gokul Rajaram, I met my teammates Kimberly Durrant and Carter Duncan. We brainstormed and proposed our ideas, suggesting the main plan and some backup plan just in case we pivot.
We ended up with the AI Business: Autonomous Website Builder Business. The AI will handle from lead generation to building websites to outreach. Clients can access their demo sites as soon as they read their emails.
We divided work and I volunteered myself on building the AI Orchestration Layer and the infrastructure of our whole codebase. I began by dumping the constraints, the prompts, the requirements, our idea to Claude Cowork. It gave us a good PRD to begin with based on what I fed it. I switched over to Claude Code and used Opus 4.6 1M Context (lifesaver, btw). It almost one-shot the initial PRD. I then optimized the whole pipeline by adding more agents that are aligned with our AI Business flow without even running the CLI. Yes, CLI. I asked it to begin by writing the prototype as a CLI. Thanks Peter.
What's interesting is that either Kimberly or Carter suggested to start work on separate directories inside the repository. I didn't know how it was going to work out, but I believed in my teammates. Kimberly built the website builder agent and Carter built the outreach agent. Both agents were written in Python. Mind you, I wrote the orchestration layer, lead hunter agent, enrichment and qualifying agents, storage agents and the project structure in TypeScript so that had me nervous. Well, it had a happy ending. We found out that Opus can handle translating the Python Code to TypeScript and even integrating it to the correct project structure. The prompt I used was simple: "My teammate built a Python agent. Help me translate and integrate it to our project structure."
After integrating everything within the project, we had Claude Code reviewing and validating the code, we fed the API keys, and did a lot of test runs with it. It was very exciting to see that it worked. It even sent outreach emails to the leads we found.
After building the whole agentic pipeline, we built the dashboard to allow the business owner to see how their agents are performing.
Goals:
- WIN 10K! (Next time.)
- ✅ Lead how I can
- ✅ Learn and Build AI Agents
- ✅ Have Fun and Meet New People
The Numbers / Results
- Scraped 260 leads across 14 cities
- 14 categories (HVAC, plumbing, electrical, mostly businesses in the trade industry)
- Generated 250+ sites after qualification and enrichment
- Full Supabase integration
- Working Dashboard
Unfortunately, we didn't get to the Finals even if we finished our product an hour before the deadline. I was beating myself up for it afterward because I was trying to deploy it in the cloud and that ate up our time trying to record a presentable demo.
Valuable Lesson: No matter how great of a product you have, marketing/pitch/demo is as important as the building part.
Jae and Aoi won the profitable award with the same idea. Great minds think alike.
Resolution
I'm all-in on AI and this rejuvenated my excitement about building. I learned I can build something really valuable in a few hours. Thanks to Carter and Kimberly for being great teammates!
The plan moving forward is to integrate the AI architecture into the project I am currently building. I only have RAG with it and have been hesitant in adding the agent layer because I'd never done it before and a bit of fear of failure (definitely not because of paying for the credits, lol).