With huge amounts of information connected to the Internet, efficient and effective discovery of resource and knowledge using the Internet has become an imminent research issue. A vast array of networks services is growing up around the Internet and massive amounts of information is added everyday. Users can now access massive amounts of information in various forms, thereby creating an equally massive problem. This rapid growth in data volume, user base, and data diversity render Internet-accessible information increasingly difficult to be used effectively. Therefore, search for a specific information on this massive and exploding Internet information resource base becomes highly critical. In this paper we discuss the issues involved in the application of machine learning techniques to the problem of Internet-based information overload. We present a general architecture and describe how it has been instantiated in a functional system we developed. The system attempts to concurrently maximize and optimize the resource/knowledge discovery, and custimize the information to individual users. We discuss the design issues involved in the attempt to develop an evolvable architecture which can easily and inexpensively accommodate future generations of web-based systems and technologies.