The Moat

AI audio production engine turning podcasts/radio into multi-platform content. Uses Gemini 3 Flash/Pro/Image to auto-scrub ads, extract clips, write articles, and generate thumbnails—hours to minutes.

Comment

Inspiration

Jeff Scofield hosts a weekly real estate show on iHeartRadio in Rochester, NY. He's been doing this for years—hundreds of episodes—but they just sit in the archive, unused.

When we asked why he doesn't repurpose content, he said:

"I don't even know that's possible. Who's going to clean this up? There are ads, dead air, interruptions everywhere. I don't have a marketing team."

We realized Jeff represents thousands of small business owners doing local radio: realtors, lawyers, financial advisors spending $500-$5K/month on airtime with zero way to multiply that investment. They're non-technical, have no resources, and don't know content repurposing is even possible.

What it does

The Moat automates the entire post-production workflow. Users upload a radio show link, wait 8 minutes, and receive a complete "50-Pack":

  • Clean audio master (ads removed)
  • 5-7 short clips (60-90 seconds)
  • 5-7 social media posts
  • 1 SEO-optimized article (1,500+ words)
  • 5-10 branded thumbnails

Zero settings. Zero technical knowledge required.

How we built it

We use Gemini 3's tri-tier architecture on Google Cloud Platform:

  • Gemini Flash - Fast audio analysis and ad detection
  • Gemini Pro - Article writing and content extraction
  • Gemini Pro Image - Brand-consistent visual generation

The platform runs entirely on GCP (Cloud Run, Firestore, Cloud Storage, BigQuery) with automated processing pipelines. Gemini's 2M context window lets us analyze full 60-minute episodes without chunking, and Pro Image's 14-image composition capability ensures visual brand consistency.

We built this in 2-3 months using Gemini CLI as a development partner.

Challenges we ran into

Radio ad detection - Radio ads have no silence gaps. We developed a two-pass system using Gemini's native audio understanding to detect structural transitions and semantic content shifts. Improved accuracy from 62% to 95%.

Audio quality - Early clips sounded robotic. Fixed with deterministic 1.5-second padding based on human breath cycles. Validated with radio professionals (8.7/10 naturalness score).

Cost optimization - Initial costs were $3.50/episode. Reduced to $1.15/episode by routing tasks to optimal Gemini models (Flash for pattern matching, Pro for reasoning).

Brand consistency - Generic AI images looked inconsistent. Used Gemini Pro Image's multi-image capability (logo + headshot + style references) to achieve 0.91 brand similarity score.

Accomplishments that we're proud of

Week → 30 minutes - Automated what previously took a week of manual editing, copywriting, and design.

Solo enterprise development - Built production-ready SaaS in 2-3 months that would traditionally require a full team.

Blue ocean market - Found 4,000-5,000 underserved radio broadcasters ignored by all "podcast AI tools."

Technical innovation - Proved hybrid architecture (AI intelligence + deterministic execution) achieves 89.5% reliability vs. industry standard 70-75%.

Gemini-exclusive features - Native audio understanding, 2M context window, 14-image composition, and MCP hooks make this impossible to build on other platforms.

What we learned

Gemini 3 is irreplaceable - Native audio processing, 2M context window, Pro Image's compositional depth, and MCP hooks are unique capabilities competitors can't match.

Target non-technical users - Jeff doesn't want features, he wants magic. Zero-config automation beats powerful configurability.

Content backlogs matter more - Users care more about unlocking 347 archived episodes than processing new ones.

Vertical focus wins - Deep solution for 5,000 radio hosts beats shallow solution for 500,000 podcasters.

Hybrid beats pure AI - Combining neural intelligence with deterministic execution outperforms AI-only approaches.

What's next for The Moat

Launch (Feb-Mar 2026) - Production deployment, initial customers, market validation

Automation (Q2 2026) - Auto-feed integration, content calendar scheduling, one-click distribution

Partnerships (Q3 2026) - Radio CMS integrations (Futuri, WideOrbit), station group white-label

Expansion (Q4 2026) - Video support, multi-language, live show processing

Long-term - Build the content operating system for 4,000+ service professionals investing in broadcast media but lacking production resources.

Built With

  • antigravity
  • cloud-functions
  • doe-architecture
  • ffmpeg
  • firestore
  • gemini-3.0-flash
  • gemini-3.0-pro
  • gemini-3.0-pro-image
  • gemini-cli
  • github
  • google-cloud-run
  • google-cloud-storage-(gcs)
  • iam-&-service-accounts
  • model-context-protocol-(mcp)
  • next.js
  • node.js
  • offlineaudiocontext
  • protocol
  • remotion
  • typescript
  • web-audio-api
Share this project:

Updates