Agent Development Kit Hackathon with Google Cloud 2025

Author

Tin Ho Chu

---

🎥 Demo Video

[Add your demo video link here]

---

🚀 Project Overview

I'm building ClipLab — a full-stack AI studio crew for content creators, powered by Google’s Agent Development Kit (ADK). This project helps YouTubers, TikTokers, and Instagram creators go from content idea to ready-to-post video scripts, captions, and trending sounds in under a minute.

Burnout is real for digital creators. ClipLab solves this by automating the pre-production process with a crew of specialized AI agents that collaborate behind the scenes.

All it takes is a 3-sentence content brief. ClipLab handles the rest — scripting, captioning, and trend hunting — giving creators more time to focus on recording and growing their audience.

---

🧠 Features & Functionality

• One-line idea input → full content pack (script, caption, trending music)
• AI agents built with ADK handle the heavy lifting:
  • 🎬 script_agent: Writes a 60s platform-native script
  • 💬 caption_agent: Generates caption + hashtags
  • 🎵 music_agent: Pulls trending TikTok sounds via tool
  • 📦 content_packager_agent: Packages it all into a ready-to-use JSON response
• Optional expansion: scheduling, analytics, shot planning

---

🛠️ Technology Stack

Frontend

• Next.js
• TailwindCSS
• Shadcn UI

Backend

• FastAPI
• Celery
• MongoDB
• Redis
• Google ADK (Agent Development Kit)

APIs & AI Models

• Google Gemini
• Google Custom Search API
• TikTok Music API (custom wrapper)

---

📡 Installation & Setup

1. Clone the repo
2. Set up your .env file (refer to .env.example)
3. (Optional) Start the frontend: npm run dev

---

🧠 Inspiration

Creators are overwhelmed by the daily pressure to ideate, write, and stay on-trend. I wanted to build an AI-powered team that could take care of everything before hitting record.

---

⚙️ How We Built It

• Multi-agent architecture using Google ADK primitives
• Each agent runs in a Celery worker to simulate async team collaboration
• Tools like Google Search and TikTok APIs power external data
• FastAPI for routing and testing pipelines
• Frontend interface built in Next.js with Tailwind and Shadcn UI

---

❗ Challenges I Ran Into

• Designing clear, prompt-safe interactions between agents
• Handling voice/audio uploads for transcription and idea extraction
• Ensuring agent outputs chained correctly without human input
• Making TikTok data accessible and usable through a tool interface

---

✅ Accomplishments That I'm Proud Of

• Fully working multi-agent pipeline using ADK
• Hands-free content pack generation from a single user prompt
• Integration of AI tools, scripting models, and live trend data

---

📚 What I Learned

• How to architect multi-agent workflows using ADK
• Prompt engineering for sequential creativity (script → caption → music)
• Real-world deployment challenges with Celery and FastAPI
• Designing AI tools that empower creators rather than replace them

---

🔮 What's Next for ClipLab

• Add post_scheduler_agent to auto-publish via Buffer/Meta/YouTube Studio
• Add shot_planner_agent to break scripts into visual scenes
• Voice input → transcription → idea pipeline
• Spanish language support for LATAM creators
• Build out user dashboards for content history and analytics

---

📝 License

This project is licensed under the MIT License. See the LICENSE file for details.