PatrickKalkman

def encode_intro_segment_core(
    video_file_path: str,
    duration: float = 300.0,  # 5 minutes
    video_height: int = 120,   # Low resolution
    video_crf: int = 28,       # Quality setting
    video_fps: int = 5,        # Reduced frame rate
) -> Optional[str]:
    """Extract and encode the first 5 minutes of the video."""
    
    # Build optimized ffmpeg command

PatrickKalkman

# Configurable quality settings
video_height = int(os.environ.get("VIDEO_HEIGHT", "120"))  # Lower = faster
video_crf = int(os.environ.get("VIDEO_CRF", "28"))         # Higher = smaller files  
video_fps = int(os.environ.get("VIDEO_FPS", "5"))          # Lower = fewer frames

# These settings reduce file sizes by 95%+ while maintaining detection accuracy 

PatrickKalkman

MOVIE_ANALYSIS_TOOLS = [
    get_video_duration,      # Determine total video length
    encode_intro_segment,    # Create optimized intro clip  
    encode_outro_segment,    # Create optimized outro clip
    detect_intro_end_time,   # AI analysis of intro segment
    detect_outro_start_time, # AI analysis of outro segment
    save_analysis_results,   # Structured result storage
]

# The agent orchestrates these tools automatically

PatrickKalkman

{
  "intro_end_time": "01:10",
  "outro_start_time": "09:50",
  "total_cost": 0.29665749999999996,
  "analysis_timestamp": "2025-05-25T12:49:02.741502",
  "video_file": {
    "name": "tears_of_steel.mp4",
    "path": "/Users/patrickkalkman/projects/st/credit-scout/tests/test_data/sample_video.mp4",
    "size_bytes": 185765954
  }

PatrickKalkman

def detect_film_start(self, video_file: types.File) -> Dict[str, float]:
    """Detect when the main film content begins."""
    
    prompt = """Analyze this video clip from the beginning of a film. 
    Identify the exact timestamp (in MM:SS format) where the first scene 
    of the main, continuous narrative body of the film begins. 
    
    This point must occur after the full conclusion of ALL of the following:
    - All studio logos and distributor cards
    - The main title card of the film itself  

PatrickKalkman

def encode_outro_segment_core(
    video_file_path: str,
    duration: float = 600.0,  # 10 minutes
    # ... other parameters
) -> Optional[str]:
    """Extract the final 10 minutes using ffmpeg's -sseof flag."""
    
    cmd = [
        "ffmpeg", "-y",
        "-sseof", f"-{duration}",  # Start from end minus duration

PatrickKalkman

Final Detailed Specification

Overview

• Purpose:
  Develop a Node.js application that runs once per day, retrieves events from a PostgreSQL database older than a configurable threshold (default one year), groups them by their UTC-created date, and exports each group as a CSV file. Each CSV is named in the format events_YYYY-MM-DD.csv and stored in an Azure Data Lake under the folder titan-pulse-event-archive. After successful upload, the corresponding events are deleted from the database in one transaction.

• Deployment Environment:
  The application will run in a Docker container (using a base image node:20-alpine3.18) deployed as a Kubernetes CronJob scheduled daily at 02:00 AM UTC in an Azure Kubernetes Cluster.

PatrickKalkman

PatrickKalkman

PatrickKalkman

resources:
  requests:
    cpu: 1500m
    memory: 1024Mi
  limits:
    cpu: 8000m
    memory: 2524Mi