Chatbots aren’t new, and they’re not always flashy. However, chat-based AI is on the rise, and Intel’s AI Assistant Builder, formerly known as Project SuperBuilder, intends to simplify the process of building and training a custom, small on-device AI assistant. I promise I’ll get to why this is exciting for you in just a moment.
AI Assistant Builder would allow creators to generate lighter LLMs for specific tasks, without dragging around all of the dead weight that heavier models are burdened with — bringing fast and relevant responses to users even when they’re offline.
The idea of creating your own AI assistant sounds incredibly complex, and there’s probably a reason that companies like OpenAI spend billions per year on training new models. However, for AI chatbots much smaller in scope, Intel has compacted the process into a single website. You simply need to specify your use case, give SuperBuilder time to create a model, download it, and launch the program.
It could revolutionize how OEMs provide technical support, help you do away with those lengthy car manuals, and add a touch of human-like flair to just about any product. Can you imagine an AI talking you through your next IKEA build? We can, and we think that’s just wild enough to be considered brilliant.
MWC 2025 Best AI Debut: Intel AI Assistant Builder (Project SuperBuilder)
Intel’s new AI Assistant Builder helps create AI assistants in minutes – YouTube
Watch On
Intel’s SuperBuilder was designed as an enterprise software solution, offering a way for various Intel partners to roll out smart features and AI Assistants with a simplified development process. The AI Assistant Builder programs can swap between LLM models like Phi and Llama, though SuperBuilder AI Assistants can also run custom models.
Additional information and reference data can be uploaded as a RAG (retrieval-augmented generation) if the AI assistant needs more specified knowledge on particular subjects. SuperBuilder AI Assistants can learn from chat history, related documents, and the inherent knowledge of the LLM. Intel has included some safety guardrails on these AI Assistants, like profanity filters and anti-bias measures.
Once rolled out to user systems, the AI Assistant can function as a highly specialized Chatbot that can also link to the RAG source material if users need more information or want to check the accuracy of the AI’s information.
Various Intel partners have already rolled out solutions using versions of Intel SuperBuilder, including Acer, Kyobo, and Asus.
One of the Intel AI Assistant Builder projects was a digitized version of a manual, allowing users to ask questions that would be answered by the user manual that no one reads.
Having spent plenty of hours setting up other people’s dive computers to the right Nitrox gas mix on dive trips, the idea of training an AI Assistant to memorize all of the various dive computer manuals for me to offer offline help in navigating the often difficult dive computer menu trees is the kind of AI innovation that would actually solve a problem. Perhaps just a very specialized one, but a problem near and dear to my own heart.
