Content Generation

Generate podcast content using AI (Anthropic Claude) with RAG-enhanced prompts from the Pinecone knowledge base.

Generate PRF

Generate a PRF (Podcast Repurposing Framework) document from a transcript.

POST /api/ai/prf

curl -X POST https://production.youvebeenheard.com/api/ai/prf \
  -H "Content-Type: application/json" \
  -H "Cookie: auth_token=<token>" \
  -d '{
    "transcript": "Full episode transcript...",
    "systemPrompt": "Generate a PRF...",
    "useRAG": true
  }'

Request

transcript

string

required

Full episode transcript (max 100,000 characters)

systemPrompt

string

required

System prompt with generation instructions

useRAG

boolean

default:true

Enable RAG context from Pinecone knowledge base (brand guidelines, writing style)

ragQuery

string

Custom RAG query (defaults to first 500 chars of transcript)

Response

{
  "content": "# Podcast Repurposing Framework\n\n## Core Message\nChris Pacifico explains..."
}

content

string

Generated PRF document (Markdown format)

RAG Context

When useRAG: true, the system:

Generates embedding for ragQuery (or transcript excerpt)
Queries Pinecone for relevant context from:
- brand-manual - YBH brand guidelines
- anti-ai-writing-guide - Writing style rules
- brand-narrative - Brand voice examples
Appends context to system prompt

Example RAG context appended:

═══════════════════════════════════════════════════════════════════
KNOWLEDGE BASE CONTEXT (Use for reference)
═══════════════════════════════════════════════════════════════════

[Brand Manual - Voice]
We don't sell. We unsell. Anti-spin, Anti-transactional, Pro-IT leader.

---

[Anti Ai Writing Guide - Avoid]
Don't use phrases like "In today's rapidly changing world" or "Let's dive in".

Model Configuration

Uses Claude Sonnet 4 (claude-sonnet-4-20250514):

Max tokens: 8,192
Temperature: Not specified (defaults to 1.0)
System prompt: Enhanced with RAG context

Generate Viral Hooks

Generate viral social media hooks from PRF.

POST /api/ai/hooks

curl -X POST https://production.youvebeenheard.com/api/ai/hooks \
  -H "Content-Type: application/json" \
  -d '{
    "prf": "PRF document...",
    "systemPrompt": "Generate viral hooks...",
    "useRAG": true
  }'

Request

prf

string

required

PRF document (Markdown or text)

systemPrompt

string

required

System prompt with hook generation instructions

useRAG

boolean

default:true

Enable brand voice context from Pinecone

Response

{
  "content": "<h2>Hook 1: The Hidden Cost</h2>\n<p>Most IT leaders...</p>"
}

content

string

Generated viral hooks (HTML format with TipTap-compatible structure)

Fact Verification: Best practice is to include the original transcript in the PRF input to enable fact-checking:

const prfInput = `SOURCE TRANSCRIPT (for fact verification):\n${transcript}\n\n═══...═══\nPRF ANALYSIS:\n${prf}`

Generate Infographic Spec

Generate infographic design specifications with layout, colors, and content structure.

POST /api/ai/infographic

curl -X POST https://production.youvebeenheard.com/api/ai/infographic \
  -H "Content-Type: application/json" \
  -d '{
    "selectedText": "Text to visualize...",
    "episodeContext": "Episode 385: Chris Pacifico...",
    "history": ["Doom Loop", "Quadrant Matrix"],
    "designType": "infographic",
    "systemPrompt": "Design an infographic..."
  }'

Request

selectedText

string

required

Text excerpt to visualize

episodeContext

string

required

Episode metadata and context (episode number, guest, transcript excerpt)

history

string[]

required

Previously used infographic types (for variety)

designType

string

default:"infographic"

Design type: infographic, quote, or thumbnail

systemPrompt

string

required

System prompt with design instructions

useRAG

boolean

default:true

Enable design database context from Pinecone

Response

{
  "spec": {
    "layout": "Pyramid (Bottom-Up)",
    "template": "Strategic Framework",
    "title": "The Three Pillars of IT Leadership",
    "colorSystem": "Corporate Professional",
    "iconStyle": "isometric",
    "aspectRatio": "16:9",
    "contentBreakdown": {
      "mainMessage": "IT leaders need three core skills",
      "sections": ["Technical Expertise", "Business Acumen", "People Leadership"],
      "dataPoints": ["70% of CIOs", "3-5 years average tenure"]
    },
    "prompt": "Create a professional isometric infographic..."
  }
}

spec

object

Infographic design specification

Show spec object

layout

string

Layout structure (e.g., “Pyramid”, “Quadrant Matrix”, “Timeline”)

template

string

Content template type (e.g., “Strategic Framework”, “Doom Loop”)

title

string

Infographic title (under 10 words)

colorSystem

string

Color palette (e.g., “Corporate Professional”, “High Contrast”)

iconStyle

string

Icon style: isometric or flat2d

aspectRatio

string

Aspect ratio: 16:9, 1:1, 9:16, etc.

contentBreakdown

object

Content structure and data points

prompt

string

Complete Kie.ai generation prompt

Design Database Context

When useRAG: true, queries Pinecone for:

Layout structures (28 types): Circular, Linear, Comparison, Grid, etc.
Content templates (10 types): Doom Loop, Strategic Framework, Journey Map, etc.
Visual style systems: Icon styles, color palettes, typography rules

Skips RAG for designType: "quote" (doesn’t need layout templates).

Default Specs by Design Type

Infographic
Quote Card
Thumbnail

{
  "layout": "Card Grid",
  "template": "Strategic Framework",
  "title": "Key Insight",
  "colorSystem": "Corporate Professional",
  "iconStyle": "isometric",
  "aspectRatio": "16:9",
  "prompt": "<selectedText>"
}

{
  "layout": "hero_quote",
  "template": "Quote Card",
  "title": "Quote Card",
  "colorSystem": "dark_cinematic",
  "iconStyle": "flat2d",
  "aspectRatio": "1:1",
  "prompt": "<selectedText>"
}

{
  "layout": "thumbnail",
  "template": "YouTube Thumbnail",
  "title": "Thumbnail",
  "colorSystem": "dark_cinematic",
  "iconStyle": "flat2d",
  "aspectRatio": "16:9",
  "prompt": "<selectedText>"
}

Error Responses

All content generation endpoints return consistent error responses:

Status	Error	Description
`400`	Transcript is required	Missing required input
`400`	PRF is required	Missing required PRF input
`400`	Selected text is required	Missing infographic input
`500`	Anthropic API key not configured	Missing API key
`500`	Anthropic API error	Upstream API error
`500`	Failed to generate [content]	Generation error

Rate Limiting & Costs

API Costs: Content generation uses the Anthropic Claude API, which incurs costs per request:

PRF generation: ~$0.10-0.30 per episode (8K tokens)
Hooks generation: ~$0.05-0.15 per set (4K tokens)
Infographic specs: ~$0.03-0.08 per spec (2K tokens)

Monitor usage via Anthropic dashboard.

No Built-in Rate Limiting: Content generation endpoints do not enforce rate limits. Implement client-side rate limiting or request throttling as needed.

Best Practices

1. Include Source Transcript

For fact accuracy, always include the original transcript:

// PRF with transcript for context
const prfResponse = await fetch('/api/ai/prf', {
  body: JSON.stringify({
    transcript: fullTranscript,
    systemPrompt: prfPrompt
  })
})

// Hooks with transcript for fact-checking
const hooksResponse = await fetch('/api/ai/hooks', {
  body: JSON.stringify({
    prf: `SOURCE TRANSCRIPT:\n${fullTranscript}\n\n═══...═══\nPRF:\n${prfContent}`,
    systemPrompt: hooksPrompt
  })
})

2. Use RAG for Brand Consistency

Keep useRAG: true (default) to inject brand guidelines:

{
  "useRAG": true,  // Fetches brand voice, writing rules, design patterns
  "ragQuery": "viral hooks for IT leaders"  // Custom query for better context
}

3. Track Generation History

For infographics, pass previous designs to ensure variety:

const history = episode.generatedAssets.map(a => a.spec.layout)

const spec = await generateInfographic({
  history: history,  // ["Doom Loop", "Timeline", "Pyramid"]
  // ... other params
})

4. Validate AI Output

Always validate generated content before saving:

if (!prfContent || prfContent.length < 500) {
  throw new Error('PRF generation failed or too short')
}

if (!spec.prompt || spec.prompt.length < 100) {
  throw new Error('Invalid infographic spec')
}

Endpoints

Services

Content Generation

Content Generation

Generate PRF

Request

Response

RAG Context

Model Configuration

Generate Viral Hooks

Request

Response

Generate Infographic Spec

Request

Response

Design Database Context

Default Specs by Design Type

Error Responses

Rate Limiting & Costs

Best Practices

1. Include Source Transcript

2. Use RAG for Brand Consistency

3. Track Generation History

4. Validate AI Output

Build docs developers (and LLMs) love

Endpoints

Services

​Content Generation

​Generate PRF

​Request

​Response

​RAG Context

​Model Configuration

​Generate Viral Hooks

​Request

​Response

​Generate Infographic Spec

​Request

​Response

​Design Database Context

​Default Specs by Design Type

​Error Responses

​Rate Limiting & Costs

​Best Practices

​1. Include Source Transcript

​2. Use RAG for Brand Consistency

​3. Track Generation History

​4. Validate AI Output

Build docs developers (and LLMs) love

Content Generation

Generate PRF

Request

Response

RAG Context

Model Configuration

Generate Viral Hooks

Request

Response

Generate Infographic Spec

Request

Response

Design Database Context

Default Specs by Design Type

Error Responses

Rate Limiting & Costs

Best Practices

1. Include Source Transcript

2. Use RAG for Brand Consistency

3. Track Generation History

4. Validate AI Output