PodGrowth - AI Podcast Growth Operator (Clips, Transcripts, Show Notes, Promotion Automation)

Acceptance Criteria

Upload: done when MP3 file uploads to S3 without errors. Transcription: done when Deepgram returns full transcript with speaker labels in under 5 minutes. Show Notes: done when Claude generates 200-word show notes with 3 key takeaways. Clips: done when 5 video clips (15–60 sec each) are extracted and playable. Approval UI: done when user can preview clips and toggle publish status. Publishing: done when approved clips send data to Zapier correctly.

How to build it, step by step

1. Bootstrap project: npx create-next-app@latest podgrowth --typescript --tailwind --app, then npm install @deepgram/sdk @anthropic-ai/sdk ffmpeg-static fluent-ffmpeg @aws-sdk/client-s3 stripe @supabase/supabase-js. 2. Supabase schema: create tables — users (id, email, stripe_customer_id), episodes (id, user_id, title, s3_audio_key, status enum[uploaded,transcribing,processing,done]), transcripts (id, episode_id, raw_json, plain_text), clips (id, episode_id, start_sec, end_sec, s3_clip_key, approved bool), show_notes (id, episode_id, content, seo_title, keywords). Enable Row Level Security on all tables scoped to auth.uid(). 3. S3 upload: create a Next.js API route /api/upload that generates a presigned PUT URL via @aws-sdk/client-s3 PutObjectCommand — client uploads directly to S3, then calls /api/episodes to create the episode row with status='uploaded'. 4. Deepgram async transcription: POST to api.deepgram.com/v1/listen?model=nova-2&diarize=true&punctuate=true&paragraphs=true with the S3 audio URL as source. Set callback_url to your /api/webhooks/deepgram endpoint. On webhook receipt, parse alternatives[0].paragraphs and store raw JSON + plain_text in transcripts table, update episode status='processing'. 5. Claude show notes + clip timestamps: POST to Anthropic API with claude-haiku-4-5, system prompt: 'You are a podcast producer. Given a transcript, return JSON with: summary (200 words), key_takeaways (array of 3 strings), seo_title (60 chars), keywords (array of 5), clip_segments (array of {start_sec, end_sec, hook_line} for 3-5 high-value moments 30-90 seconds each).' Parse response and insert into show_notes and clips tables. 6. FFmpeg clip extraction: create a Node.js worker script (not a serverless function — run on a VPS or AWS Lambda with 5-min timeout) that reads clip rows with no s3_clip_key, downloads the source audio from S3 using signed GET URL, runs fluent-ffmpeg with .seekInput(start_sec).duration(duration_sec).audioCodec('aac').toFormat('mp4') to produce a video-ready clip (black background + waveform optional via ffmpeg lavfi), uploads output to S3, updates clip row with s3_clip_key. 7. Approval UI: build /dashboard/episodes/[id] page in Next.js App Router — fetch clips via Supabase client, render each clip with an HTML5 audio player (use signed S3 GET URL), show hook_line as caption preview, toggle approved status via PATCH /api/clips/[id]. Show show_notes and social post templates (pre-filled from Claude output) with copy buttons. 8. Social post templates: in the Claude prompt from step 5, also request social_posts: {twitter: string (280 chars), linkedin: string (1200 chars), threads: string (500 chars)} — display these in the approval UI as editable text fields. 9. Zapier webhook trigger: on user clicking 'Publish Approved', POST to a Zapier Catch Hook URL with payload {clip_url, caption, platform[]}. User sets up Zap to route to YouTube/Twitter/LinkedIn. Document this in onboarding. 10. Stripe billing: create products in Stripe dashboard (Pro $29, Creator $79, Studio $199). Use stripe.checkout.sessions.create with mode:'subscription' on /api/billing/checkout. Handle stripe webhook events customer.subscription.created and customer.subscription.deleted to update users.plan in Supabase. Gate upload form by checking plan + episode count for current billing period.