In this product demo, Christopher Parsons walks through the Revit Assistant, a knowledge agent built into the Synthesis platform for AEC firms. He explains the UI action buttons that launch pre‑canned Revit queries, the concept of grounding knowledge (using Revit‑tagged posts, the Revit course catalog, and a Revit standards guide) as the exclusive data source for the agent, and the human fallback/contact routing feature that directs unanswered queries to a designated expert (e.g., Director of Design Technology Brianne Walton). Parsons demonstrates a live voice‑mode interaction where the assistant diagnoses a slow Revit model by pulling performance tips from its grounded knowledge, citing specific source pages and recommending a relevant learning course. He also showcases conversation history, rating, and feedback tools that let users flag inaccuracies or missing knowledge, feeding back into the knowledge management process.
Chapters
Okay, so now I'm in the Revit Assistant. This is a knowledge agent within Synthesis, and I'm gonna tell you some kind of basics about how all agents work. And so we'll take a little bit longer kind of unpacking the Revit Assistant. So here are what we call action buttons. So in this example, these are kind of pre-canned queries for like commonly asked questions about, you know, Revit that the Revit Assistant can answer. So if you click on one of those, it'll execute and launch the prompt and it will go answer that question. Um, but that also serves to kind of like advertise to users like what the capabilities of this agent are. So a couple key concepts to get into. One of them is something called grounding knowledge. And so grounding knowledge is the knowledge from within Synthesis that the agent is allowed to use to answer questions. So in this example, we have grounded it in the Revit hashtag. So these are any posts within the platform that are used Revit hashtag. So it can use those to answer questions. We have grounded it in the course catalog and filtered it on the Revit courses. So it just uses the Revit courses. And then we've grounded it in our Revit standards guide. So this is in our technology community, a lot of information about terminology, how we set up projects, the best way to build models, et cetera. So that knowledge and that knowledge alone is what the Revit Assistant is allowed to use to answer questions. Okay. One more thing I want to show you while we're here. So if it gets out beyond its knowledge base, we have a feature called the human fallback contact. So if it can't answer a question, you can configure it not to have a human fallback contact. You can have it suggest employees that it should contact if it can't answer the question, or you can have it suggest an email or a link. So it could be, you know, help@yourfirmname to your support ticket, or it could be some kind of other portal that you want to send people to. In this example, Brianne Walton is our Director of Design Technology. So it's routing questions to Brianne if the Revit Assistant can't answer the questions. Okay. You can see that I've asked some, um, in, in conversation history, you can see that I'm able to go through recent conversations I've had with the agent. And now I'm just gonna start a new one and I'm gonna use voice mode for this. My Revit model's running pretty slow. Do you have any recommendations on how I can make it faster? Great. So now it's gonna go through, it's gonna look through, um, that grounding knowledge that we talked about before for performance tips. And now it's gonna start answering that. And you can see that as it's answering, it's citing the sources in the underlying grounding knowledge. So for example, it's citing this model upkeep page as where it got its information. And here you can see it's also citing citing a general config, uh, considerations page. It's citing project setup, model construction, and then it also at the bottom references a Learning Center course called Revit Model Health and Best Practices and encourages you to go take that course. So you can continue to go on and ask follow-up questions if you don't understand what any of these things are. You can rate your conversation and give feedback to the knowledge management team or the company in terms of This was inaccurate, or it wasn't able to answer my question because we're missing some knowledge, or I don't understand why it gave me the answer. So a lot of tools for helping to keep improve the knowledge built straight into the agent platform.