South Park Commons x Anthropic Hackathon Demos

Key Concepts

Anthropic Hackathon, South Park Commons (SPC), Large Language Models (LLMs), Tool Calling, AI Agents, Design Thinking, Product Development, Software Development, User Interface (UI), Web Scraping, E-commerce, Sales, Pitching, APIs, MCP (Multi-Chain Protocol) Server, Shopify, Mobile App Development, Git Integration, Workflow Automation.

Hackathon Overview and Introduction

The YouTube video transcript documents the demo night of a hackathon hosted by South Park Commons (SPC) in collaboration with Anthropic. The event showcased projects built over the weekend using Anthropic's Claude models and other technologies. Ten finalist teams presented their projects to a panel of judges, competing for prizes including Best Overall, Most Interesting Interface, 10x Engineering, and Audience Choice.

Project Demonstrations

1. Hacka Babies: AI Bartender

• Concept: A physical tiki bar with an AI bartender powered by Claude.
• Functionality: The AI engages in a conversation with the user to determine their preferences, mood, and allergies. Based on this, it designs a custom cocktail using available ingredients. The system uses an LLM to interact with an Arduino board to dispense the drink in the correct proportions.
• Technical Details: LLM acts as a bartender persona, using a tool to control eight peristaltic pumps for precise ingredient mixing.
• Example: Crystal has a conversation with the AI, which recommends a "midnight reviver" based on her mood.
• Notable Quote: "Greetings professor Mix here share your mood and I'll create the perfect cocktail" - AI Bartender
• Outcome: Won the Audience Choice Award. The physical tiki bar was being auctioned off.

2. Bongo Browser: Tactile Web Interface

• Concept: A tactile interface for web browsing using physical objects (Diet Coke cans) as input devices.
• Functionality: The system uses a language built around the browser (LSD) to interpret taps on the cans as clicks on the webpage. Claude is used to decide what to click on based on the sanitized HTML of the page.
• Technical Details: The language takes sanitized HTML, sends it to Claude, Claude decides what to click on, and the language executes that on the browser.
• Example: Tapping an aluminum foil on a can triggers a click on a random element on a Wikipedia page.

3. Team Kanto: PokePlay.ai - AI Pokemon Trading Card Game Player

• Concept: An AI agent that can play the Pokemon Trading Card Game.
• Functionality: The system uses Langraph for orchestration and Langchain for individual agents. A player agent makes moves based on the game state (represented as a JSON object). A mentor agent advises the player agent, and a referee agent validates the legality of the moves.
• Technical Details: Orchestration is done through langraph, individual agents are made through lang chain.
• Example: Ash (the AI player) uses a Battle VIP pass to get Ralts and Mew onto the bench and attaches a psychic energy to Zashian VI.
• Key Argument: The team aimed to create an AI agent that could enable millions of children around the world to learn and play the Pokemon Trading Card Game.
• Outcome: Won the Most Interesting Interface award.

4. Claude Code Go: Mobile Coding App

• Concept: A mobile app that allows developers to code on the go using Claude.
• Functionality: The app connects to a desktop Git repository, allowing users to make changes, preview code, and create pull requests from their phone.
• Technical Details: The app connects to a live Git repository and allows users to chat with Claude, preview changes, and manage Git operations.
• Example: The presenter uses the app to make the dragon size oscillate and add fire emojis to a website.
• Outcome: Won the 10x Engineering prize.

5. Team Paid: Paid Product AI Design Agent

• Concept: A conversational AI agent that guides users through the design thinking process.
• Functionality: The agent asks questions based on IDO and Stanford Design School design thinking principles to help users define their product, target audience, and value proposition. It generates a Notion document, a mermaid script for visualizing the user flow, and Excalidraw wireframes based on the conversation.
• Technical Details: The agent is trained on design thinking principles and uses Claude to generate artifacts based on the conversation transcript and a JSON representation of the design thinking process.
• Example: Eli has a conversation with the agent about building a surf app, and the agent generates a user flow and wireframes based on the conversation.

6. Team Nardwire: AI-Powered Research Agent

• Concept: An AI agent that mimics the research style of interviewer Nardwuar.
• Functionality: The agent researches individuals based on calendar invites and email context, generating in-depth profiles with interesting facts and meeting preparation questions.
• Technical Details: The agent uses a multi-agent architecture with Anthropic 3.7 Sonnet, tool calling, web APIs, and LinkedIn/X scraping.
• Example: The agent researches the judges of the hackathon and generates meeting preparation questions.
• Outcome: Won the Best Overall prize.
• Notable Quote: "We basically took the re that used claude to build a researcher to mimic exactly what Nardoir would do" - Team Nardwire

7. Team Yellow Pages: Nonprofit Resource Validation

• Concept: An AI agent that validates contact information for nonprofit organizations.
• Functionality: The agent scrapes websites to find and validate email addresses for aging and disability resource centers. It provides a screenshot to help volunteers verify the information.
• Technical Details: The agent uses computer vision and natural language processing to identify and validate email addresses on websites.
• Example: The agent validates the email address for the Aging and Disability Resource Center in Autonom, Washington.

8. Team A Z: Web Client for MCPs

• Concept: A web client for Multi-Chain Protocols (MCPs) that simplifies their use for non-technical users.
• Functionality: The client provides templates for common use cases (e-commerce, operations) and visual diagrams to represent complex workflows.
• Technical Details: The client uses a compiler LLM to generate programs from conversation traces and represents workflows as flowcharts.
• Example: The presenter demonstrates a workflow for listing products on Shopify, finding the most expensive item, and sending a summary to Slack.

9. Team Pitchcraft: AI-Powered Sales Coach

• Concept: An AI-powered sales coach that helps technical founders improve their pitching skills.
• Functionality: The platform provides feedback on company blurbs and sales decks, generates customer questions and sales outreach emails, and allows users to practice their pitch with AI feedback. It also connects users with human sales coaches.
• Technical Details: The platform uses AI to analyze pitch content and provide feedback based on best practices.
• Example: The presenter has the AI analyze a drunk pitch by Don Draper.

10. Team New Generation: AI Storefront for Brands

• Concept: An AI storefront for brands that allows customers to interact with products through chat.
• Functionality: The system aggregates product data from various brands and exposes it through an MCP server. Customers can ask questions about products in chat, and the system can notify merchants of customer interest.
• Technical Details: The system uses an MCP server to connect to product data and allows for two-way conversations between consumers and merchants.
• Example: A customer asks about buying commemorative trading cards of the hackathon judges and expresses interest in having one made of themselves.

Prize Categories

• Best Overall: Awarded to the team with the most impressive project overall (Team Nardwire).
• Most Interesting Interface: Awarded to the team with the most creative and engaging user interface (Team Kanto).
• 10x Engineering: Awarded to the team that built the best tool for developers (Team Claude Code Go).
• Audience Choice: Awarded to the team that received the most votes from the audience (Team Hacka Babies).

Conclusion

The Anthropic hackathon showcased a diverse range of innovative projects leveraging Claude and other technologies. The projects demonstrated the potential of AI to solve problems in various domains, including bartending, web browsing, gaming, coding, product development, research, nonprofit resource validation, e-commerce, and sales. The event highlighted the importance of user-friendly interfaces, practical applications, and the integration of AI with existing workflows.