Summary of "Start your Cloud AI project: Practical Demos and your Options at Google Cloud"
Summary of the Subtitles (Technological Concepts & Google Cloud AI Options)
The session explains how to start a Cloud AI project on Google Cloud, presenting multiple paths depending on skill level and role. It includes practical demos using Google Cloud credits (Trail Billing), hands-on API calls with Postman, and ML model creation using BigQuery ML, Vertex/Agent Platform Workbench (Jupyter Lab/Notebook), and AutoML.
1) Two Main Categories: “Ready-Made” vs “Build Your Own Model”
Google Cloud offers two broad approaches:
- Ready-made AI solutions / turnkey services (no model-building required)
- Build your own ML models (with options that simplify the process)
2) Option A — AI Solutions (Turnkey / Ready-Made Services)
Demo concept: a contact-center / virtual agent style service for a telecom company.
- Users interact through a UI (virtual agent/assistant)
-
The agent can call backend systems to perform actions (e.g., top up plans, increase data allowance)
-
Human customer service staff can intervene
- Uses a knowledge base and can handle new user cases with insights
3) Option B — Machine Learning APIs (Integrate Into Applications)
For developers integrating AI into applications, Google provides prebuilt APIs, including:
- Text & Speech APIs (speech-to-text and text-to-speech)
- Translation API
- Text classification / sentiment analysis / understanding
- Vision APIs (Cloud Vision / Vision Intelligence) for images and video
Demo: Cloud Vision API Face Detection
The demo uses a “try the API” workflow:
- Upload an image using a Try the API link
- The API detects multiple faces and returns structured results including:
- Face bounding regions (rectangle coordinates)
- Facial keypoints (e.g., eyes, eyebrows, nose, mouth, chin)
- Detection outputs appear as structured response objects/arrays
4) Hands-On: Getting Google Cloud Credits (Trail Billing) + Project/Billing Setup
A major portion of the session is a step-by-step guide to enable trial credits so participants can try APIs.
Key requirements and steps:
- You need a Google Cloud Project before using services (VMs, storage, ML, etc.)
- Trial billing credits are tied to a Billing Account
- The billing account must be linked to the project
- The session includes instructions for:
- signing into the correct registered account
- updating “working links” (e.g., last number changes)
- creating/switching projects in the Google Cloud Console
- linking the correct billing account to the project
Reminder: ensure the billing account name like “Cloud Platform Trail Billing” is used for the project.
5) Hands-On: Calling Vision API via Postman (Authentication + Request Body)
The demo shows how to request Cloud Vision through Postman.
Request setup
- Use POST requests to the Vision API endpoint
- Authentication options mentioned:
- API key
- token-based method
- service account (for server-to-server / service identity)
API key security restrictions (important)
- Restrict the key to the correct API (Cloud Vision API)
- Restrict by application type (web/iOS/Android) and possibly IP addresses
Postman specifics
- Set the API key in Postman request authorization
- Build JSON request bodies based on Google documentation examples
- Use an image reference (URL) and set parameters like:
facedetectionmaxResults(example shown:2)
Response explanation
The response includes an array of detected faces. Each face entry can contain:
- Bounding polygon / coordinates (x,y points)
- Confidence and detection details
- Facial landmark keypoints (e.g., left/right eye/eyebrow, nose, mouth, etc.)
6) Object Storage Workflow for Feeding Images to Vision API
Instead of relying on external image URLs, the session demonstrates uploading an image to Google Cloud Storage (object storage), then calling Vision API with the stored object.
Concepts covered:
- Use object storage so applications can scale without depending on app server/web URLs
- Create a Storage bucket (called “pack” in the subtitles)
- Choose:
- location (single region / multi-region; example: US multi-region)
- storage class (Standard, Archive, Auto mentioned)
- Discuss public access when needed for demo access
- Upload an image (example: a rose image)
- Use the resulting public object URL in the Vision API request
- Vision API returns results such as label detection (example label: “flower”)
7) Option C — Create ML Models with BigQuery ML (Analyst/Data Scientist Path)
For cases where data is in SQL/data warehouses, Google Cloud provides BigQuery ML.
- Use BigQuery for large-scale analysis (GB/TB/PB)
- Demo uses a public dataset (Google Analytics-related tables)
- SQL-based workflow includes:
- querying across date-suffixed tables
- defining a label for classification:
- purchase occurred or not (transaction vs no transaction)
- engineering features such as:
- operating system / platform
- page visit counts
- handling a country field
Model training and prediction
- Create a model using logistic regression for binary classification (buy vs not buy)
- Use the trained model via
ML.PREDICTto predict outcomes for new/unseen time periods - Post-processing:
- filter predictions by country
- aggregate and count expected buyers
8) Option D — Custom ML Engineering with Vertex/Agent Platform Workbench (Jupyter Lab)
For ML engineers building custom pipelines:
- Use Agent Platform (subtitles refer to it as the “new name for Vertex CE”)
- Use Workbench to launch JupyterLab/Jupyter Notebook
- Demo includes:
- cloning a GitHub repository
- running notebook workflows
Emphasis for large data handling:
- connect the notebook to BigQuery instead of loading everything into notebook memory
- mentions BigFrames conceptually as a data-frame connector to BigQuery
9) Option E — AutoML for Beginners (No Deep ML Expertise)
AutoML helps create models without extensive ML knowledge.
- Uses transfer learning with a wizard-style experience
- Demo workflow includes:
- create a dataset
- choose task type (example includes image classification and tabular classification)
- define labels/target columns
- upload data (CSV) from:
- Cloud Storage or BigQuery/other sources
- choose training region and run settings
- specify training budget/time
- After training, produce a model ready for deployment via endpoints
10) Closing: Five Options Recap + Encourage Using Credits
The session ends with a recap of five practical pathways:
- AI Solutions (ready-made)
- Machine Learning APIs for app integration
- BigQuery ML for analysts/data scientists building models in SQL
- Custom ML with Workbench (Jupyter Lab) for engineers
- AutoML for beginners / wizard-driven model creation
The speaker encourages participants to:
- use the trial credits
- try the APIs/ML workflows hands-on during the workshop
Main Speakers / Sources Mentioned
- Fadi Nabil (trainer / workshop presenter)
- Engineer Ghada (ITI support; acknowledged during introductions)
- Google for Developers and ITI (Information Technology Institute) (event collaborators)
- Google Cloud / Google Cloud services demonstrated: Cloud Vision API, Cloud Storage, BigQuery ML, Agent Platform/Workbench, AutoML
Category
Technology
Share this summary
Is the summary off?
If you think the summary is inaccurate, you can reprocess it with the latest model.