Check if required packages are installed.

Checks if the requested period overlaps with existing rosters. Args: start_date: Start date in YYYY-MM-DD format num_days: N

Classify a list of sampled frames using parallel processing. Args: sampled_frames: List of sampled frame bytes. client: Gemini cl

Classify rotation with detailed results. Returns dict containing: - classification: The final classification - reason: Why this c

Stop and remove the container with proper cleanup.

Compute features from image embeddings.

Strip HTML of tags and add separators to convert observation into simple mode

Create the orders table with proper schema.

Create BigQuery table with time partitioning if it doesn't exist.

Async wrapper for Deep Research to prevent blocking the main app loop. Includes a timeout and conciseness injection for performance.

Describe all garment images in parallel. Args: client: Gemini client instance garment_images: List of garment images as bytes or

Create an ANN index on the collection's embedding field if it doesn't exist. Without an index, semantic search is brute-force and stops working p

Evaluate a generated background change image against reference face. Uses shared process pool from person_eval to avoid DeepFace threading issues

Evaluate a generated VTO image against reference face using InsightFace. Args: vto_image_bytes: Generated VTO image bytes referen

Call Gemini Pro to surgically correct specific fields in the output. Unlike execute() which regenerates the entire Postprocessing

Extract high-level summary from agent outputs.

Extract sections from the rules book relevant to the failing phase/step. Rather than sending the entire 80K rules book to the LLM, extrac

Queries HUD CHAS API for housing problem and cost burden statistics. Args: county_fips: 5-digit County FIPS code or common city

Fetches HUD Income Limits (AMI) with nested JSON schema matching. Args: county_fips: 5-digit County FIPS code or common city nam

Queries HUD USPS crosswalk API to map ZIP code to County FIPS code (type=2).

Queries MLS listings for a target metropolitan area and performs automated investment analysis (Cap Rate, Cash-on-Cash Return) by correlatin

Find files that were deleted from GCS but still tracked in BQ. Compares the preprocessed table (our known state) against the Object Table (li

Identify which steps are completed and which is next (pure function).

Formats the DDL schema for use in SQLGlot.

Find the range of frames that have the same class as either the first or last image. Uses binary search, searching only within 60 frames from the

Call Gemini Pro and return (response_text, latency_ms).

Orchestrates the entire deck generation process.

Generates an image and saves it to the specified path. Returns image bytes on success.

Generates a single image using Gemini, handling retries. Args: prompt (str): The prompt for image generation. input_images (List[

Generate videos using Veo, automatically handling batching for requests > 4 videos. Args: client: The Veo client collage_img: The

* Retrieves a customer based on their ID. * * @param currentCustomerId The ID of the customer to retrieve. * @returns The Customer object if

Instantiates the Auditor Judge agent using ADK.

Discover all evalset files and their test files.

Returns the set of readers for a value and the visited objects. Args: value: The value to get the readers for. visited_objects: The s

Get list of available LangChain Gmail tools.

Simulates getting the current time for a city. Args: query: The name of the city to get the current time for. Returns: A str

Retrieves a customer based on their ID. Args: customer_id: The ID of the customer to retrieve. Returns:

Fetches raw log entries from Google Cloud Logging for Dataproc resources. This is a common helper function that constructs a filter string t

Executes a gcloud command to fetch and wait for (or stream) logs for a traditional Dataproc Job or Batch, returning the raw command result.

Generate queries to set up the BigQuery dataset and tables.

Get or create a timestamped folder for this debug session.

Extract face embedding from an image using InsightFace. Args: img_bytes: Image as bytes. mask_eyes: If True, mask the eye regions

Standardize entity ID: Austin (48453) -> 4845399999.

Get or create the InsightFace FaceAnalysis app. Uses buffalo_l model for highest accuracy, running on CPU via ONNX.

Provides product recommendations based on the type of plant. Args: plant_type: The type of plant (e.g., 'Petunias', 'Sun-loving annuals')

Gets the GCP project ID from environment or gcloud config.

Get all active rules for a specific scope.

Fetches secret from Secret Manager efficiently.

Heuristic to map city names to their primary states.

Get tolerance configuration for a specific step. Returns: Dict with: - value: The tolerance value

Build user agent string identifying this agent.

Lazy-load the webshop environment on first access without using global.

Creates a Dataproc WorkflowTemplateServiceClient for a specific region. Args: region: The GCP region for the client. Returns:

Initializes and returns a Gemini client. Returns: A genai.Client instance or None if initialization fails.

Complete investigation of agent processing for one case. v2.0.0: Added per_group_validator and discovered_rules for Layer 3 validation.

Render and return the HTML for a product item page

Generates a Python function that evaluates a policy query using Vertex AI.

Loads a CSV file into a BigQuery table. Args: data_project_id: GCP Project for BQ data. dataset_name: The name of the BigQuery da

Load JSON file

Load JSON file safely

Loads metadata from a GCS URI.