Project
WebWisp
Autonomous distributed web testing agent
An autonomous web agent for end-to-end website testing with distributed workers, task orchestration, and browser-grounded execution.
WebWisp is an autonomous web agent developed during my internship at LaBRI. It uses a distributed architecture for scalable web testing and automation:
Orchestrator
Dispatches tasks to worker pools
Exposes a monitoring interface for runs and outcomes
Handles authentication and user session control
Workers
Execute browser tasks reliably at scale
Support horizontal scaling across multiple containers
Use Playwright for deterministic browser automation
Integrate WebSOM for precise element grounding
Message broker (RabbitMQ)
Guarantees asynchronous communication between services
Decouples orchestration from execution for scalability
The system is containerized with Docker and includes JWT-based auth, persistent storage, configurable worker pools, and LLM-assisted testing flows.
This implementation is informed by advances in in-context language modeling and visual grounding for web interfaces.
An autonomous web agent for end-to-end website testing with distributed workers, task orchestration, and browser-grounded execution.
WebWisp is an autonomous web agent developed during my internship at LaBRI. It uses a distributed architecture for scalable web testing and automation:
Orchestrator
Dispatches tasks to worker pools
Exposes a monitoring interface for runs and outcomes
Handles authentication and user session control
Workers
Execute browser tasks reliably at scale
Support horizontal scaling across multiple containers
Use Playwright for deterministic browser automation
Integrate WebSOM for precise element grounding
Message broker (RabbitMQ)
Guarantees asynchronous communication between services
Decouples orchestration from execution for scalability
The system is containerized with Docker and includes JWT-based auth, persistent storage, configurable worker pools, and LLM-assisted testing flows.
This implementation is informed by advances in in-context language modeling and visual grounding for web interfaces.
Related Experience
References
- [2020] Language Models are Few-Shot Learners — Brown, T. B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., Neelakantan, A., Shyam, P., Sastry, G., Askell, A., Agarwal, S., Herbert-Voss, A., Krueger, G., Henighan, T., Child, R., Ramesh, A., Ziegler, D. M., Wu, J., Winter, C., Amodei, D.. arXiv preprint arXiv:2005.14165 DOI ↗
- [2023] Set-of-mark prompting unleashes extraordinary visual grounding in GPT-4V — Yang, J., Zhang, H., Li, F., Zou, X., Li, C., Gao, J.. arXiv preprint arXiv:2310.11441 DOI ↗