Python SDK Quickstart

Integrate SeeMe.ai into your Python applications. This guide covers installation, authentication, and common operations.

What You’ll Accomplish

Install and configure the SDK
Authenticate with your account
Create datasets and upload data
Train a model
Make predictions programmatically

Prerequisites

Python 3.8 or higher
pip package manager
A SeeMe.ai account (sign up)

Step 1: Install the SDK

pip install seeme

Verify the installation:

python -c "import seeme; print(seeme.__version__)"

Step 2: Initialize the Client

from seeme import Client

## Create a client instance with your credentials
client = Client(username="your-username", apikey="your-api-key")

ℹ️

You can store your credentials in a .env file with username, apikey, and optionally backend variables. Then simply call Client() without arguments.

Alternative: Environment File

For convenience, create a .env file in your project directory:

username=your-username
apikey=your-api-key

Then initialize without arguments:

client = Client()

Step 3: Create a Dataset

from seeme.types import Dataset, DatasetContentType

# Create a new image dataset
my_dataset = Dataset(
    name="My Training Dataset",
    description="Images for classification training",
    content_type=DatasetContentType.IMAGES,
    default_splits=True
)

my_dataset = client.create_dataset(my_dataset)

print(f"Dataset created: {my_dataset.id}")

Step 4: Upload Data

Upload a ZIP file

# ZIP structure: folder name = label
# my-data.zip/
#   cats/cat1.jpg, cat2.jpg...
#   dogs/dog1.jpg, dog2.jpg...

# Get the first version
version = client.get_dataset_versions(my_dataset.id)[0]

client.upload_dataset_version(
    dataset_id=my_dataset.id,
    dataset_version_id=version.id,
    folder=".",
    filename="my-data.zip"
)

Upload individual files

from seeme.types import Label, DatasetItem, Annotation

# Get the default version and splits
versions = client.get_dataset_versions(my_dataset.id)
version = versions[0]
splits = client.get_dataset_splits(my_dataset.id, version.id)
train_split = splits[0]

# Create labels
cat_label = client.create_dataset_label(
    dataset_id=my_dataset.id,
    dataset_version_id=version.id,
    label=Label(
        name="cat",
        color="#FF5733",
        version_id=version.id
    )
)

dog_label = client.create_dataset_label(
    dataset_id=my_dataset.id,
    dataset_version_id=version.id,
    label=Label(
        name="dog",
        color="#3498DB",
        version_id=version.id
    )
)

# Upload an image
item = client.create_dataset_item(
    dataset_id=my_dataset.id,
    dataset_version_id=version.id,
    item=DatasetItem(
        name="cat1.jpg"
    )
)

client.upload_dataset_item_image(
    dataset_id=my_dataset.id,
    dataset_version_id=version.id,
    item_id=item.id,
    folder=".",
    filename="cat1.jpg"
)

# Add a label annotation
client.annotate(
    dataset_id=my_dataset.id,
    dataset_version_id=version.id,
    annotation=Annotation(
        item_id=item.id,
        split_id=train_split.id,
        label_id=cat_label.id
    )
)

Step 5: Train a Model

from seeme.types import Job, JobType, JobItem, ValueType, Framework, ApplicationType

# Get the application_id for your framework
application_id = client.get_application_id(
    base_framework=Framework.PYTORCH,
    framework=Framework.FASTAI,
    base_framework_version="2.1.0",
    framework_version="2.7.13",
    application=ApplicationType.IMAGE_CLASSIFICATION
)

# Create a training job
my_job = Job(
    name="Train cat vs dog classifier",
    job_type=JobType.TRAINING,
    application_id=application_id,
    dataset_id=my_dataset.id,
    dataset_version_id=version.id,
    items=[
        JobItem(name="nb_epochs", value="10", value_type=ValueType.INT, label="Number of epochs"),
        JobItem(name="image_size", value="224", value_type=ValueType.INT, label="Image Size"),
        JobItem(name="batch_size", value="32", value_type=ValueType.INT, label="Batch Size"),
        JobItem(name="arch", value="resnet50", value_type=ValueType.TEXT, label="Architecture"),
    ]
)

my_job = client.create_job(my_job)
print(f"Training started: {my_job.id}")

Monitor Training Progress

import time
from seeme.types import JobStatus

while True:
    my_job = client.get_job(my_job.id)
    print(f"Status: {my_job.status}")

    if my_job.status in [JobStatus.FINISHED, JobStatus.ERROR]:
        break

    time.sleep(30)

if my_job.status == JobStatus.FINISHED:
    print(f"Model created: {my_job.model_id}")

Step 6: Make Predictions

from seeme.types import ApplicationType

# Get your model
model = client.get_model(my_job.model_id)

# Make a prediction
result = client.predict(
    model_id=model.id,
    item="./test-image.jpg",
    application_type=ApplicationType.IMAGE_CLASSIFICATION
)

# Display results
for item in result.inference_items:
    print(f"{item.prediction}: {item.confidence:.2%}")

Batch Predictions

import os

# Predict on multiple images
image_folder = "./test-images/"
for filename in os.listdir(image_folder):
    if filename.endswith(('.jpg', '.png')):
        result = client.predict(
            model_id=model.id,
            item=os.path.join(image_folder, filename),
            application_type=ApplicationType.IMAGE_CLASSIFICATION
        )
        top = result.inference_items[0]
        print(f"{filename}: {top.prediction} ({top.confidence:.2%})")

Complete Example

Here’s everything in one script:

from seeme import Client
from seeme.types import (
    Dataset, DatasetContentType, Job, JobType, JobItem,
    ValueType, Framework, ApplicationType, JobStatus
)
import time

# Initialize
client = Client(username="your-username", apikey="your-api-key")

# Create dataset
my_dataset = Dataset(
    name="Cat vs Dog Classifier",
    content_type=DatasetContentType.IMAGES,
    default_splits=True
)
my_dataset = client.create_dataset(my_dataset)

# Upload data (assumes ZIP with cats/ and dogs/ folders)
version = client.get_dataset_versions(my_dataset.id)[0]

client.upload_dataset_version(
    dataset_id=my_dataset.id,
    dataset_version_id=version.id,
    folder=".",
    filename="training-data.zip"
)

# Get the application_id
application_id = client.get_application_id(
    base_framework=Framework.PYTORCH,
    framework=Framework.FASTAI,
    base_framework_version="2.1.0",
    framework_version="2.7.13",
    application=ApplicationType.IMAGE_CLASSIFICATION
)

# Train model
my_job = Job(
    name="Train cat vs dog",
    job_type=JobType.TRAINING,
    application_id=application_id,
    dataset_id=my_dataset.id,
    dataset_version_id=version.id,
    items=[
        JobItem(name="nb_epochs", value="10", value_type=ValueType.INT, label="Number of epochs"),
    ]
)
my_job = client.create_job(my_job)

# Wait for training
while True:
    my_job = client.get_job(my_job.id)
    if my_job.status in [JobStatus.FINISHED, JobStatus.ERROR]:
        break
    time.sleep(30)

# Make prediction
if my_job.status == JobStatus.FINISHED:
    result = client.predict(
        model_id=my_job.model_id,
        item="./test.jpg",
        application_type=ApplicationType.IMAGE_CLASSIFICATION
    )
    for item in result.inference_items:
        print(f"{item.prediction}: {item.confidence:.2%}")

Common Operations Reference

Task	Method
List datasets	`client.get_datasets()`
Get dataset	`client.get_dataset(id)`
Delete dataset	`client.delete_dataset(id)`
List models	`client.get_models()`
Get model	`client.get_model(id)`
List jobs	`client.get_jobs()`
Delete job	`client.delete_job(id)`
Make prediction	`client.predict(model_id, item)`

What’s Next?

Full SDK Reference Build Something Automate Workflows

Video Script Outline

Hook (0:00-0:10) “Add AI to your Python app in under 10 minutes. Let’s build something.”

What You’ll Learn (0:10-0:30)

Install the seeme package
Authenticate and create a client
Upload data and train a model
Make predictions in your code

Demo Steps (0:30-7:00)

pip install, show version check (0:30)
Create client, show auth flow (1:00)
Create dataset programmatically (1:45)
Upload ZIP, explain structure (2:30)
Start training job (3:30)
Monitor progress loop (4:30)
Make prediction, show output (5:30)
Show batch prediction example (6:30)

Call to Action (7:00-8:00) “You now have a working AI pipeline in Python. Check out the full SDK docs at docs.seeme.ai for advanced features.”

No-Code Quickstart REST API Quickstart