03Architecture
01 / Documents
/upload accepts product knowledge for ingestion.
OPENING THE SYSTEM
Loading the next page…
Loading the next page…
Turning product knowledge into test scenarios, test cases and executable Selenium scripts.
A retrieval-augmented pipeline ingests domain knowledge, retrieves context and uses an LLM to generate test scenarios, cases and Selenium scripts.
Product documents describe expected behaviour, but that knowledge must be translated into concrete scenarios before it becomes test automation.
Feature: account sign-in Expected behaviour: A user with valid credentials can access their dashboard.
/upload accepts product knowledge for ingestion.
Generated tests are bounded by the ingested product knowledge. Script generation does not establish that a test is correct or safe to run.
Application documentation, retrieved document chunks, test scenarios and structured test cases.
Separate ingestion, retrieval, test generation and script generation into inspectable API stages. Reuse the same contextual knowledge throughout the pipeline.
Asynchronous Python and FastAPI expose /upload, /query, /testcase and /script. The retrieval stack combines all-MiniLM-L6-v2 embeddings with ChromaDB; Gemini and LangChain support generation.
A staged workflow translates documents into retrieved context, scenarios, test cases and Selenium scripts, keeping intermediate outputs inspectable.
No measured test coverage, pass rate or generation accuracy has been supplied. Generated scripts should be reviewed and evaluated against the actual application.
The system architecture connects domain knowledge to executable automation. The portfolio demonstration illustrates the workflow; it does not execute an agent or browser test.
Retrieving sufficient context, avoiding unsupported test assumptions and translating expected behaviour into robust selectors and assertions.
Inspectable intermediate representations make generated test automation easier to reason about and review.
Add a benchmark of grounded scenarios, validate generated scripts in an isolated environment and document failure-repair behaviour.
The repository link has not been published. Ask about the implementation