Hackathon Overview

Topic: Develop a business model using generative AI APIs / Develop a business model using generative algorithms

Our team’s choice: Develop a business model using generative AI APIs

When the notice was posted on the university website, it described the event as a university alliance hackathon with meals and snacks provided during a two-night, three-day stay at a hotel. Since my exams had just ended, I thought it would be a good opportunity to rest and work on a development project at the same time.

The organizers said teams would be assigned as fairly as possible based on the Google Form submitted in advance. In total, five universities participated, and each team had five members.

Our team first started generating ideas, and the following candidates came up.

1. Chrome extension that summarizes terms of service
2. Gamification -> fairy tale implementation & psychological test -> Gamification
3. Book summarizer
4. Emotion diary
5. Prompt converter -> formats prompts for AI [Markdown, multiple forms]
6. Generative AI that can bypass GPT detectors
7. Learning curriculum for major students
   -> further developed
8. Phone notification summarizer by importance + automatically removes alerts when the user is looking at the phone (Face)
   -> further developed
9. AI that provides solutions to real-world problems
10. AI for analyzing part-time jobs, occupations, and reviewing contracts
11. Programming learning
12. No More Vibe Coding & Quiz AI
13. Code -> Tagging AI
14. Summarize AI conversation history + prompt -> continue into the next conversation
15. Baby-language argument robot
    -> further developed
16. Room-cleaning recommendation AI
    -> further developed
17. AI detector
18. Information reliability verification AI
19. Drone frequency location verification AI
20. Wearable + AI for optimal sleep-time analysis
21. Food photo analysis + healthcare AI

Team Composition and Roles

I took the role of project lead. The team consisted of one frontend developer, one backend developer, one designer, and one hardware designer.

The roles were divided as follows.

• Project Lead (me): Overall planning and development direction, technical architecture design, schedule and role coordination, PPT and presentation leadership, related paper and market research
• Frontend (1): UI development and React-based conversation page implementation
• Backend (1): FastAPI server setup, STT/TTS/LLM API integration
• Design (1): Character design, logo design, and PPT
• Hardware (1): Raspberry Pi 5 design, sensor, microphone, and speaker module composition using online tools

Idea: A Baby-Language Argument Robot Toy

The idea was inspired by https://youtu.be/Tgk9L-6XsvA, with the additional image of a moving cactus toy. It was selected through a team vote.

While researching this idea, we found that “babbling” does not carry a fixed semantic meaning, which made it unsuitable for direct implementation using a generative AI API.

As we considered changing the idea, we found another point from the research: language ability grows rapidly in children between around 18 months and 3 years old. So instead of throwing away the idea completely, we shifted the focus to children aged 18 months to 3 years and decided to develop it more concretely.

Data: https://talkbank.org/ | https://langcog.github.io/childes-db-website/index.html Existing product: https://catius.cafe24.com/

Revised Idea: An Interactive Robot Toy That Grows With the Child

Core idea:

1. Grow with the child -> provide developmental-stage information and required abilities for the relevant age range through our app.
2. Let the child talk with the toy and provide reports that help check language and cognitive abilities according to the developmental stage.
3. Provide age-appropriate conversation content based on developmental-stage theory, such as 18 months to 3 years, 3 to 4 years, and 4 to 4.5 years.
4. Support language and brain development.
5. Design hardware and a cute exterior that can withstand impacts when a child throws or hits it.

Problems?

1. How can a very young child operate the toy easily? -> Use camera recognition, make the switch visually obvious, etc.
2. GDPR-K in Korea and regulations around children’s online services
3. Algorithmic error and interpretation risk -> fallback behavior, model diversity in training [problem: insufficient data]
4. Technology and regulation

Project Structure

Tech Stack

Frontend: React, Vite, Tailwind Backend: FastAPI, WebSocket, Python Gen AI: GPT-4-turbo, Clova STT Database: SQLAlchemy + Alembic for user/log storage Deployment: localhost

Problem-Solving Process

(Focused on key issues such as errors, ports, DB, deployment, and AI communication)

Result and Demo

(Frontend-backend communication success, MVP completion)

Retrospective and Takeaways

(Sense of achievement, lessons learned, and next goals)

Links

(GitHub, demo URL, etc.)