Yohan Raju works on getting current web data into AI applications: reaching pages, extracting useful fields and keeping collection working as websites change. His work connects web-data infrastructure with practical uses ranging from training-image and video datasets to ecommerce intelligence and agents that need fresh information to answer a question.
Yohan's professional background includes New York Institute of Technology and front-end and back-end information technology. In 2024, he was a solutions consulting specialist at Bright Data, co-leading a web-data acquisition workshop with Rafael Levi. His subsequent demonstrations connect that collection work with AI applications whose inputs change after a model has been trained.
Making collection work beyond the prototype
Yohan treats collection as a pipeline extending beyond the initial request. Pages must be discovered, changing selectors and different layouts handled, browser and proxy infrastructure managed, and returned fields validated before records reach downstream applications. Finding a relevant website is only the beginning of supplying an application with usable evidence.
His conversation with Ramon Ray places this work in a broader business context: collecting image and video datasets for training, gathering ecommerce intelligence, and helping organizations scale beyond their internal collection resources. Integrating web-data infrastructure into existing pipelines is central to that account. A team with an established collection process can add infrastructure where its own resources become insufficient, rather than rebuilding the entire workflow.
Connecting collection to an agent's task
Yohan demonstrates integrations with Bright Data APIs and Bright Data's MCP server, which exposes tools for web search, page extraction and browser interaction. His contribution here is explaining and demonstrating their use; the server is a Bright Data project.
One product-intelligence example combines an Amazon listing's price, rating and product details with Reddit discussion, then passes both to Gemini for an explained purchase recommendation. The example gives fresh, structured evidence a specific purpose: product records supply the listing's details, while customer discussion adds sentiment. It connects Yohan's broader collection work to the decision an application is meant to help someone make.
Yohan Raju follows web data from fragile selectors and access barriers to a product intelligence notebook that combines an Amazon listing, Reddit discussion and Gemini analysis.
Web collection has separate access, extraction, validation and delivery responsibilities. A page can load successfully while a changed selector silently removes a required field.
Freshness requirements should follow the question. Current-year travel advice may need new evidence; repeated queries can reuse cached information while it remains adequate.
The notebook combines Amazon product fields with Reddit discussion before asking Gemini for a recommendation. Obtaining fresh inputs does not by itself establish that the resulting sentiment or demand claims are representative.
▶ Watch ↗