Reverb

Logo

🔥 Inspiration

Creators spend hours converting long-form content: podcasts, interviews, videos - into posts for LinkedIn, Twitter/X, and Reddit. While clipping tools exist, nothing focuses on text-first repurposing where writing style and structure actually drive engagement.

Reverb was built to automate this process - without sacrificing your voice.

💡 What It Does

Reverb helps creators repurpose long-form content into platform-optimized posts... in seconds.

Step 1: Upload a YouTube video or audio file
Step 2: Provide a few writing samples (X posts, captions, etc.)
Step 3: Reverb transcribes, learns your tone, and generates:
- 🐦 X / Twitter: Hook-driven threads with CTAs and hashtags
- 👔 LinkedIn: Structured, spaced-out insights
- 🔍 Reddit: Conversational posts for authentic engagement

Everything sounds like you, not a generic AI assistant.

🛠 How We Built It

Frontend: React + TailwindCSS
Backend: FastAPI (Python)
Transcription: OpenAI Whisper
Content Generation: OpenAI GPT-4 with structured prompting
Style Matching: Custom tone extractor + prompt conditioning
RAG System: ChromaDB + pgvector for retrieval from user samples
Storage: Supabase (media + metadata)

We used ChromaDB to implement a lightweight RAG (Retrieval-Augmented Generation) pipeline that grounds generation in the creator’s actual past posts; enabling voice consistency across all outputs.

🚧 Challenges We Ran Into

Matching tone without fine-tuning, while still sounding human
Keeping content structured and coherent across long transcripts
Making threads that actually read like threads, not just lists
Building a clean UI for users to preview and interact with output

🏆 Accomplishments We're Proud Of

Built a full pipeline: upload → transcription → tone detection → platform-specific generation
Maintained a recognizable, consistent tone across all platforms
Ran a blind A/B test with 40 users and over 500 comparisons - Reverb’s output was preferred over baseline ChatGPT 80% of the time
Delivered "wow" moments from users: “This feels like something I’d write”

📚 What We Learned

Great AI UX comes from structured prompt flows, not just model power
RAG is shockingly effective for voice alignment, even without model training
Creators don’t want summaries - they want repackaged, ready-to-post content
One smooth, focused workflow is better than five disjointed features