Transform client call recordings into actionable insights. Extract key details, action items, and follow-ups from video calls without manual review.

​
Features

  • Video Processing: Upload Loom files or analyze YouTube URLs
  • Smart Summarization: Automatic extraction of highlights and action items
  • Custom Prompts: Use default templates or create custom analysis workflows
  • Fast Processing: Powered by Gemini 2.0 Flash API

​
How It Works

The analyzer processes videos through a structured pipeline: 
Video Upload → Context Extraction → LLM Analysis → Insight Generation → Storage

​
Core Components

ComponentPurpose
Gemini Files APIVideo upload and validation
Gemini Cache APIContext extraction and preprocessing
LLM EngineAnalysis using GPT-4 or Gemini models
Prompt TemplatesStructured analysis instructions
Token TrackerUsage monitoring and cost calculation

​
Implementation

​
Video Analysis Pipeline

def analyze_video(video_url, custom_prompt=None):
    # 1. Upload and validate video
    file_response = gemini_files_api.upload(video_url)
    
    # 2. Extract context via cache API
    context = gemini_cache_api.process(file_response)
    
    # 3. Initialize LLM with context
    llm = initialize_llm(model="gemini-2.0-flash")
    
    # 4. Apply analysis prompts
    prompts = load_prompt_templates(custom_prompt)
    
    # 5. Generate insights
    response = llm.analyze(context, prompts)
    
    # 6. Store results
    return store_response(response)

​
Processing Flow

  1. Upload Validation
    • Validates file size and duration constraints
    • Checks format compatibility
  2. Context Extraction
    • Pre-processes video using Gemini Cache API
    • Extracts semantic context for analysis
  3. LLM Analysis
    • Applies prompt templates for sales-focused insights
    • Generates structured output with key findings
  4. Result Storage
    • Saves analysis to backend storage
    • Tracks token usage and costs

​
API Reference

​
Endpoints

Primary Analysis Endpoint 
POST /analyze-video
Request Body 
{
  "videoLink": "https://loom.com/share/example",
  "promptType": "default|custom",
  "customPrompt": "Focus on objection handling and engagement",
  "clientContext": {
    "name": "Acme Corp",
    "project": "Q4 Demo Review"
  }
}
Response Format 
{
  "summary": "Client showed strong interest in automation features",
  "actionItems": [
    "Follow up with pricing proposal by Friday",
    "Schedule technical deep dive for next week"
  ],
  "keyInsights": [
    "Budget concerns around implementation timeline",
    "Strong alignment with current workflow needs"
  ],
  "tokenUsage": 1250,
  "processingTime": "3.2s"
}

​
Functions

FunctionPurpose
getAIProAgentChatResponse()Standard LLM agent processing
getSalesCallResponse()Sales-specific analysis pipeline

​
Configuration

​
Model Settings

  • Primary Model: Gemini 2.0 Flash (recommended)
  • Alternative: GPT-4 for specialized analysis
  • Context Window: Up to 1M tokens

​
File Constraints

  • Maximum Duration: 2 hours
  • Supported Formats: MP4, MOV, WebM
  • File Size Limit: 500MB
  • Source Types: Direct upload, Loom URLs, YouTube URLs

​
Analysis Templates

Built-in templates for common analysis patterns:
  • Objection identification and handling
  • Sentiment and engagement tracking
  • Talk ratio and conversation flow
  • Key decision points and next steps

​
Usage Examples

​
Basic Analysis

curl -X POST /api/analyze-video \
  -H "Content-Type: application/json" \
  -d '{
    "videoLink": "https://loom.com/share/abc123",
    "promptType": "default"
  }'

​
Custom Analysis

curl -X POST /api/analyze-video \
  -H "Content-Type: application/json" \
  -d '{
    "videoLink": "https://youtube.com/watch?v=xyz789",
    "promptType": "custom",
    "customPrompt": "Focus on technical questions and feature requests"
  }'

​
Performance Considerations

​
File Limits

  • Maximum video duration: 2 hours
  • Supported formats: MP4, MOV, WebM
  • File size limit: 500MB

​
Processing Time

  • Average processing: 30-60 seconds per hour of video
  • Parallel processing supported for multiple videos
  • Caching reduces repeat analysis time by 80%

​
Error Handling

Common error scenarios and responses:
Error CodeDescriptionSolution
VIDEO_TOO_LARGEFile exceeds size limitCompress video or split into segments
INVALID_URLURL not accessibleVerify video is public and URL is correct
QUOTA_EXCEEDEDAPI rate limit reachedImplement request throttling
PROCESSING_FAILEDAnalysis pipeline errorRetry with different prompt or video format

​
Integration Notes

​
Authentication

API calls require valid authentication tokens. Include in request headers: 
Authorization: Bearer your-api-key

​
Rate Limits

  • 100 requests per hour per API key
  • 10 concurrent video processing jobs
  • Token usage tracked per request

​
Webhooks

Optional webhook support for async processing: 
{
  "webhookUrl": "https://your-app.com/video-analyzed",
  "events": ["analysis.completed", "analysis.failed"]
}