Browser Automation + Local-LLM Content Generation (R&D)
Overview
An R&D build combining Selenium browser automation with locally-run language models. It scripts registration and review flows on an online classifieds platform, generates synthetic identities, and uses an offline LLM to draft text content automatically.
Why It Exists
A sandbox for two questions: how far Selenium IDE-exported flows can be parameterized and re-driven programmatically, and whether a fully local, offline LLM (GPT4All) can generate usable freeform text without any cloud API. The platform involved is an adult online-classifieds site, so the work is kept internal.
What We Built
A set of Python scripts: Selenium IDE-generated page-object classes (test_login.py, test_review.py, test_registration.py) parameterized with email, credentials and target IDs and re-driven via ChromeDriver; a faker_data.py for synthetic identity/profile data and temporary-inbox retrieval (getInbox.py); and gpt4all2.py, which loads a local GPT4All model from disk and generates/extracts text fully offline, with a separate Hugging Face stub for comparison. A flows.side Selenium IDE project ties the recorded flows together.
Technologies & Approach
Selenium IDE captured the flows; the exported Python was refactored into reusable, parameterized classes. GPT4All provided a zero-cost, no-network LLM running locally on the developer machine, useful for testing prompt-driven generation without API keys or data leaving the box.
Outcome / Impact
Demonstrated practical chops in scripted browser automation and in running offline local LLMs for text generation and entity extraction, both reusable techniques for QA automation and privacy-sensitive AI tasks.
Capabilities Demonstrated
- Selenium-based end-to-end browser automation and flow parameterization
- Re-driving recorded Selenium IDE flows programmatically
- Running fully-offline local LLMs (GPT4All) for generation and extraction
- Synthetic identity/data generation with Faker and temporary inboxes