PodCraft

• 

• 

  Homescreen for the Ui version.

Inspiration

Podcast production is broken. Independent creators spend 5-10 hours per episode juggling recording, editing, music selection, and audio mastering. Audiobook producers face similar bottlenecks. Large studios waste thousands of dollars on post-production overhead.

We built PodCraft to solve this: a multi-agent AI system that turns a simple podcast script into a complete, production-ready audio episode in minutes. Inspired by the "Agentic Cinema" theme, we imagined a world where AI crew members work together like a film production team — a Director, a Researcher, and an Audio Producer — to handle the entire podcast production pipeline.

What it does

PodCraft is a fully automated podcast production agent powered by Google Cloud's Gemini Enterprise. Here's how it works:

Upload a script — Users upload a podcast script or interview transcript in PDF format

Director Agent — Analyzes script structure, identifies speakers, detects mood and tone, and creates production notes

Researcher Agent — Uses Parallel Search API to find market intelligence, similar successful podcasts, and audience engagement trends

Audio Producer Agent — Generates multi-speaker audio using Gemini TTS (assigning unique voices to different speakers), creates mood-matched background music via Lyria 3, and performs sentiment analysis comparing script tone vs. audio delivery

Final Output — A complete, downloadable podcast episode with production recommendatio

How we built it

We built PodCraft using Google Cloud's Agent Development Kit (ADK) to orchestrate a multi-agent system: Core Technologies: Gemini Enterprise (gemini-2.0-flash-exp) — Powers all agent reasoning and decision-making Google Cloud Agent Engine — Hosts and orchestrates ADK agents in a serverless environment Parallel Search API — Provides real-time web intelligence for market research Gemini TTS — Generates natural, expressive multi-speaker audio Lyria 3 — Creates mood-matched background music based on script sentiment FastAPI — Backend API framework Docker & Cloud Run — Containerization and deployment Architecture Pattern: Director-Worker Pattern — A Director agent decomposes tasks and delegates to specialized worker agents Document Processing — PDF scripts are parsed using PyPDF2 to extract structured dialogue, speakers, and scene data ADK Native — All agents are built using Google's native ADK (not external wrapper libraries), ensuring compliance with hackathon rules

Challenges we ran into

PDF Parsing Complexity — Podcast scripts come in varied formats. We built a flexible parser that identifies speaker names, dialogue segments, and scene breaks using regex patterns.

Multi-Speaker Voice Assignment — Assigning unique TTS voices to different speakers while maintaining natural conversational flow was challenging. We implemented a speaker-to-voice mapping system that tracks each speaker's dialogue segments.

Sentiment Analysis Integration — Comparing script tone against audio delivery required careful prompting to ensure Gemini accurately captured emotional nuances.

API Rate Limits — Managing API calls across multiple agents while respecting free tier limits required careful orchestration and caching strategies.

Audio File Management — Handling large audio files in a serverless environment required streaming and temporary file management.

Accomplishments that we're proud of

✅ Full multi-agent workflow from script upload to finished audio ✅ Successful integration of 4 major APIs: Gemini, ADK, Parallel, and Lyria ✅ Production-ready Docker containerization ✅ Clean, intuitive API with FastAPI ✅ Comprehensive documentation for judges

What we learned

What's next for PodCraft

🎯 Add support for video scripts and automatic video podcast generation

Integrate real-time collaboration for multi-creator workflows Add translation support for global podcast creators Build a web-based UI for non-technical users Integrate with popular podcast hosting platforms (Spotify, Apple Podcasts)

Built With

• docker
• fast-api
• gemini-enterprise
• gemini-tts
• google-cloud
• google-cloud-agent-engine
• lyria-3
• parallel-search-api
• python
• vertex-ai-agent-development-kit