Structured Data Extraction

{
  "defect_found": true,
  "defect_type": "scratch",
  "location": "left side",
  "severity": "low"
}

defect_found: true
defect_type: scratch
location: left side
severity: low

{
    "defect_found": True,
    "defect_type": "scratch",
    "location": "left side",
    "severity": "low"
}

defect_found: YES
defect_type: scratch
location: left side
severity: low

{"defect_found": true, "defect_type": "scratch", "location": "top edge", "severity": "low"}

defect_found: true
defect_type: scratch
location: top edge
severity: low

The image shows a scratch. Here is the result: {"defect_found": true, ...}

You are a quality inspection assistant. Examine the product image and return a JSON object with the following fields:
- defect_found (boolean): true if any defect is visible, false otherwise
- defect_type (string or null): brief description of the defect, null if none
- location (string or null): where the defect appears, null if none
- severity (string): one of "none", "low", "medium", or "high"

Respond with only the JSON object, no additional text.

You are an equipment inspection assistant. Examine the image of the equipment and return a YAML document with the following fields:
- equipment_id (string): the ID visible on the equipment label
- status (string): one of "operational", "needs_maintenance", or "out_of_service"
- issues (list): each issue has type, location, and severity
- next_inspection (date): recommended next inspection in YYYY-MM-DD format

Respond with only the YAML output, no additional text or code fences.

You are a UI-to-code assistant. Given a screenshot of a web UI component, generate the HTML and inline CSS that reproduces the layout. Use semantic HTML elements. Use placeholder values for dynamic content like images and text. Output only the HTML code, no explanation.

import json
from vi.inference import ViModel

model = ViModel(
    run_id="your-run-id",
    secret_key=".your-secret-key.",
    organization_id="your-organization-id",
)

result, error = model(
    source="product_image.jpg",
    user_prompt="Inspect this product for defects.",
    generation_config={
        "temperature": 0.0,   # deterministic output for consistent structure
        "do_sample": False
    }
)

if error is None:
    raw = result.result  # string output from the model
    data = json.loads(raw)

    print(f"Defect found: {data['defect_found']}")
    if data['defect_found']:
        print(f"Type: {data['defect_type']}")
        print(f"Location: {data['location']}")
        print(f"Severity: {data['severity']}")
else:
    print(f"Inference error: {error}")

import yaml
from vi.inference import ViModel

model = ViModel(
    run_id="your-run-id",
    secret_key=".your-secret-key.",
    organization_id="your-organization-id",
)

result, error = model(
    source="equipment_photo.jpg",
    user_prompt="Inspect this equipment and report its status.",
    generation_config={"temperature": 0.0, "do_sample": False}
)

if error is None:
    data = yaml.safe_load(result.result)
    print(f"Equipment: {data['equipment_id']}")
    print(f"Status: {data['status']}")
    for issue in data.get('issues', []):
        print(f"  Issue: {issue['type']} at {issue['location']}")
else:
    print(f"Inference error: {error}")

import json
from vi.inference import ViModel

model = ViModel(
    run_id="your-run-id",
    secret_key=".your-secret-key.",
    organization_id="your-organization-id",
)

result, error = model(
    source="product_image.jpg",
    user_prompt="Inspect this product for defects.",
    generation_config={"temperature": 0.0, "do_sample": False}
)

if error is not None:
    print(f"Inference error: {error}")
else:
    try:
        data = json.loads(result.result)
        # process structured data
        print(data)
    except json.JSONDecodeError:
        # log raw output for debugging
        print(f"Could not parse output. Raw: {result.result}")

import json
from vi.inference import ViModel

model = ViModel(
    run_id="your-run-id",
    secret_key=".your-secret-key.",
    organization_id="your-organization-id",
)

image_paths = ["img1.jpg", "img2.jpg", "img3.jpg"]

results = model(
    source=image_paths,
    user_prompt="Inspect this product for defects.",
    generation_config={"temperature": 0.0, "do_sample": False}
)

records = []
for image_path, (result, error) in zip(image_paths, results):
    if error is not None:
        print(f"{image_path}: error - {error}")
        continue
    try:
        data = json.loads(result.result)
        data["image"] = image_path
        records.append(data)
    except json.JSONDecodeError:
        print(f"{image_path}: could not parse output - {result.result}")

# records is now a list of dicts ready to write to a database or CSV
print(f"Processed {len(records)} images successfully")