How It Works¶

Video Intelligence Agent turns your application demo videos into ready-to-use BDD test cases through a fully automated, AI-powered pipeline. Here is what happens from the moment you send a video to the moment you receive your feature files.

The End-to-End Flow¶

flowchart TD
    A(["You send a video"]) --> B["Video Intelligence Agent receives the request"]
    B --> C["Video is forwarded to Gemini AI"]
    C --> D["Gemini analyses all screens, flows, and interactions"]
    D --> E["BDD test cases are generated per business domain"]
    E --> F["Feature files stream back to you one by one"]
    F --> G["A summary report is delivered"]
    G --> H(["Done — your .feature files are ready"])

Step 1 — You Send a Video¶

You start by sending a video of your application to Video Intelligence Agent. The video can be:

A full screen recording of a product demo
A walkthrough of specific user flows
Any recording that shows your application in action

You can also include an optional text message alongside the video to give context, such as:

“Focus on the login and checkout flows”

Video Intelligence Agent uses this context to prioritize the areas you care about most while still covering all other visible flows.

Supported video formats

MP4 only. Videos up to size of 25 MB are supported.

Step 2 — Video Analysis¶

Video Intelligence Agent forwards your video to Google Gemini — a multimodal AI model capable of understanding both visual and textual content simultaneously.

Gemini carefully watches the video and identifies:

All application screens and pages
User workflows and navigation paths (login, checkout, registration, etc.)
Form fields, buttons, dropdowns, and other interactive elements
Success states and error states (validation messages, alerts)
Data displays such as tables, lists, and cards
Authentication and authorization flows
Any edge cases visible in the recording

Step 3 — BDD Generation¶

Using an expert QA engineering system prompt, Gemini produces structured Gherkin test cases organized by business domain (e.g., authentication, checkout, user management) — not by page or screen name.

For each domain identified, a separate .feature file is created containing:

Section	Description
Feature header	Name, user story (As a / I want / So that)
Background	Shared preconditions for all scenarios in the file
Happy path scenarios	Primary success flows
Negative scenarios	Empty fields, wrong credentials, unauthorized access
Boundary scenarios	Maximum lengths, special characters, rapid repeated inputs
Scenario Outlines	Data-driven tests with Examples tables

Step 4 — Streaming Delivery¶

Video Intelligence Agent does not wait until everything is ready before responding. Feature files are streamed back as they are generated, so you start receiving results almost immediately.

You receive the following sequence of real-time events:

Event	What it means
Task received	Your request has been accepted
Analyzing video	Gemini is processing the video
Generating feature files	Results are being assembled
Artifact per feature file	Each `.feature` file delivered one at a time
`summary.json` artifact	Final metadata report
Completed	All results have been delivered

Step 5 — Output¶

You receive one or more .feature file as per the application demo video, each containing fully tagged, production-ready Gherkin scenarios.

You also receive a summary report that includes:

Total number of feature files generated
Total number of scenarios across all files
The application name as detected from the video
A list of all user flows that were identified

What the AI Uses to Guide Generation¶

Video Intelligence Agent's AI is guided by an expert system prompt written from the perspective of a senior QA engineer with 15+ years of experience. The prompt instructs the AI to:

Write steps in business language — no technical selectors or IDs
Keep each step describing one behavior only
Ensure each scenario is fully independent
Tag every scenario consistently with priority, type, and nature
Keep scenarios focused: 3–8 steps per scenario
Use Scenario Outline over duplicated scenarios wherever possible

Tagging Convention¶

Every generated scenario is automatically tagged. Tags communicate at a glance what the scenario covers and how important it is.

Tag	Meaning
`@P1`	Critical — must pass for a release
`@P2`	Important — regression suite
`@P3`	Nice to have
`@smoke`	Core happy-path checks
`@regression`	Full regression suite
`@positive`	Testing the expected success path
`@negative`	Testing invalid inputs or error states
`@boundary`	Testing limits and edge cases
`@data-driven`	Parameterized test with Examples table