This repository contains a semi-autonomous web interaction agent built using Python and Playwright.
The agent opens a real website, observes page elements at runtime, makes decisions based on what it sees, and logs its actions and reasoning.
The goal of this project is not full automation, but to demonstrate decision-making and reasoning in uncertain web environments.
- Launches a browser using Playwright
- Observes page elements such as headings, links, and buttons
- Applies simple heuristic rules to decide next actions
- Logs observations, decisions, and failures with timestamps
A separate script (flipkart_agent.py) demonstrates how the agent behaves on a highly dynamic, real-world website like Flipkart.
Due to login popups and dynamic DOM updates, some interactions may fail. These failures are logged as insights rather than treated as errors.
- Heuristic-based decision making instead of hard-coded flows
- Focus on transparency and debuggability
- No LLM usage to keep behavior predictable and explainable
- Does not handle login or checkout flows
- Designed for analysis and reasoning, not full website automation
- LLM-based reasoning for complex decision-making
- Improved recovery strategies for dynamic content