Orchestrating Fleet Intelligence with AI Agents and Real-Time IoT With Databricks + DOMO
Summary
- Agentic Digital Twins: Build autonomous digital twins of mechanical assets and human workflows using AI agents on Databricks and Domo.
- Real-Time IoT Analysis: Ingest and analyze streaming fleet telemetry like bearing vibration data with FFT to detect anomalies automatically.
- Autonomous Workflows: AI agents diagnose root causes, simulate scenarios, identify service locations and personnel, and generate corrective strategies with human review.
Orchestrating Fleet Intelligence with AI Agents and Real-Time IoT With Databricks + DOMO

Talk by Eddie Edgeworth, CTO, Koantek (Sponsor Speaker)
In today’s logistics landscape, operational continuity depends on real time awareness and proactive decision making. this video presents an AI agent driven solution built on Databricks that transforms real time fleet IoT data into autonomous workflows. Streaming telemetry such as bearing vibration data is ingested and analyzed using FFT to detect anomalies. When a critical pattern is found, an AI agent diagnoses root causes and simulates asset behavior as a digital twin, factoring in geolocation, routing, and context. The agent then generates a corrective strategy by identifying service sites, skilled personnel, and parts, estimating repair time, and orchestrating reroutes. It evaluates alternate delivery vehicles and creates transfer plans for critical shipments. The system features human AI collaboration, enabling teams to review and execute plans. Learn how this architecture reduces downtime and drives resilient, adaptive fleet management.
Intelligent analytics for everyone: https://www.databricks.com/product/business-intelligence
BI meets AI: https://www.databricks.com/resources/ebook/business-intelligence-meets-ai
Explore the only unified platform for agent systems: https://www.databricks.com/product/artificial-intelligence
See all the product announcements from Data + AI Summit: https://www.databricks.com/events/dataaisummit-2025-announcements
FAQs
How quickly can you build this fleet management system?
This solution was built in under 40 hours using a partnership between Databricks, Domo, and partner Koantek, demonstrating the rapid deployment capability.
What types of anomalies does the system detect?
The system analyzes real-time fleet telemetry including bearing vibration data using FFT (Fast Fourier Transform) to automatically detect mechanical anomalies in assets.
How can I enable real-time business intelligence on operational data?
Real-time BI on IoT and operational data requires ingesting high-frequency events (device telemetry, fleet locations) into an analytics platform with sub-second latency. This video demonstrates how DOMO and Databricks enable fleet operators to monitor vehicle location, fuel consumption, and maintenance needs in real time. Streaming data from thousands of vehicles flows into the Databricks Data and AI platform where it's immediately queryable by dashboards and AI agents. Real-time visibility into fleet operations enables dynamic routing, predictive maintenance, and faster incident response.
What are the latency characteristics of real-time analytics on operational data?
This video demonstrates latency characteristics when combining Databricks with real-time BI tools like DOMO: IoT data ingested via Confluent streams reaches the Databricks Data and AI platform with sub-second latency, making it immediately available for BI queries and AI agent decisions. Query latency depends on aggregation complexity—simple metrics (current vehicle count, average speed) query in milliseconds, while complex aggregations over millions of events may take seconds. For most operational decisions (routing, alerts, anomaly detection), this latency is sufficient and dramatically faster than batch processing.
What are the best business intelligence tools?
The best BI tools for operational data combine real-time data connectivity, low-latency query performance, and ease of use for business teams. This video shows DOMO as an example of a modern BI platform that can connect directly to Databricks for real-time dashboards and alerting. Choice depends on your needs: self-service analytics (Databricks SQL, Looker, Tableau), embedded analytics (DOMO), or real-time operational dashboards (Databricks AI/BI). Most enterprises benefit from a combination—rich self-service BI for exploratory analysis plus purpose-built operational dashboards for real-time decisions.
Full transcript
We're going to talk quickly about um we have listened Aentic Fleet Ops, but really what we're going to talk about is how using the partnership uh between data bricks and Domo specifically with a partner KATCH in under 40 hours we were able to build a full agentic AI I driven digital twin of both a mechanical asset in a field and a human process in dispatch. So first, if you're not familiar, what is DOMO and data bricks look like?
Well, Domo extends your data brick investment. All the data that you have within uh data bricks can be leveraged and used using Domo in in place and we can layer on top. We have over 1,000 plus connectors that are already ready to go out of the box to help ingest and hydrate data lakeink but also allow you to ingest uh other data much more easily into your pipeline. We have a drag and drop no code low code tool for doing ETL and transformation that all operates on your data bricks compute.
And we have a fully contained front end that allows you to do low code, no code all the way up to procode app and visualization and interface. What this means is that we can take what we currently see as the modern analytic stack where we have our sources connectors ETL ELT engine data bricks and then apps and ML. And we can make this much more efficient. Right? Coming from the environment that I've come from over the years, the more places I have for failure, the more likely I'm going to have failure.
It also means the more experts in a system I need, which can slow down development and time to value. Domo allows you to simplify this entire data bricks powerhouse by simply going from data acquisition into data bricks with ETL and AI capability on top of data bricks layered in for you and then out to your procode and no code apps and visualizations all in an entirely uh contained and synchronous environment. No longer do you have to have multiple tools and know how to use multiple tools in order to access your data and in so summarize we can connect to the data prepare it with smarter insights and build agents faster through power smarter a agent it with comb uh com combined data uh with powerful AI and agent capabilities easily connect and prepare your data to transform it engage with real-time analytics which we're going to talk about and then workflows and automations which can drive your agents as well as just plain ain institutionalized best business practices in environments which will make sure that it happens automatically and correctly every time.
So the agent in particular we're going to talk about was a rag based agent that was utilizing the low latency capabilities of data bricks which has just been discussed and announced right so data ingested near real time from IoT devices in the field in this case telemetrics from fleet devices real time ML ops operating on that using autoML to identify the best model to determine anomalies that needed to be sent to the agent and then using SQL alert real time triggers over to an agent in Domo.
The agent in Domo then takes that data along with additional operational data and other sources whether it be SOPs, other procedures etc or in this case even vibration data around bearings and bring it together and execute multiple asynchronous interactions with humans for human in the loop. agentic driven AI solutions, right? In this case, fleet maintenance and logistics. Both of them get separate interface uh communication from the agent identifying areas of concern and actions to take and then coordinate those actions ultimately sending it back to the actual devices in the field, the fleet operators uh upon acceptance and approval.
I want to go through that step by step here. Step one, as I indicated, using the uh low latency capabilities of data bricks, everything's happening near real time. Data comes up, gets processed, analyzed, and any any event that needs to be triggered is immediately triggered. From there, Domo picks it up, takes and acts on that trigger, analyzes the data in context with the rest of the organizational data and information that's been given and takes care of action orientation.
One of the key advantages of this is why what is the beauty of the agent here? We could use just rules. You maybe could, but the problem is rules are just that rules. And they only apply when you're within the boundary of which those rules mattered. And usually those are suboptimum decisions. Those are decisions we've established that we're going to take the known downside for the security of it always being general.
In this case, what you're going to see is that by harnessing LLMs and AI in this capability, we are allowed to create dynamic, realistic, and sophisticated solutions that are meaningful to the people in the field to take better decisions and actions than they would have before. What does this look like in Domo from an agent perspective? Well, as I mentioned before, we pride ourselves in the fact that everything can be done low code, no code with the opportunity for pro code.
In this case, here's an example of low code, no code. With just dragging on shapes and setting them up, you can set up agents in minutes to hours, not weeks or months. Okay? Again, as I mentioned before, and I'll come back to again, one of the coolest things about this project when we put it together was it took us less than 40 hours, people hours, not a week of 10 people, but less than one full-time equivalent to put together the full agent that you will see here.
Another key component that you can leverage with Domo in combine in combination with data bricks is the fact that we provide a fully interactive user interface. This means that we can create not only message boarding like this out of the box, but these are clickable, they're actionable. The human being can actually enter comments, changes, decisions, whatever in there. And that data can be sent back throughout the rest of the pipeline and even sent back into data bricks based on actions chosen.
Additionally, using the front-end capabilities, we could bring a full true digital twin of the situation to bear. In this case, if you're looking at this, this might be the fleet uh maintenance board where they can actually see which component in real time has failed or is likely to fail on this particular asset. Now, we're using fleet here as an example, but anybody who has any other manufacturing or IoT uh needs understands the power of what a digital twin down to the asset level with graphical capability pulling from your own work uh work information in terms of work orders, work instructions and technical documents.
What that means to be able to do that in real time in an interface like this. Additionally, as we said, we can tailor the migration for the users. You get to choose all this You can control who sees what with what governance because we leverage data bricks as governance. You can control who sees what with what governance, what action to take and to filter those actions back through the rest of the process. The agent will wait for maintenance to determine whether or not they're going to accept the recommendation or one of the recommendations.
If so, it then executes over to the actual maintenance facility, letting them know to expect this vehicle. Meanwhile, it's waiting for dispatch to approve if there were any high priority packages or things that were perishable that mattered that needed to get there now that could not survive that delay to be in order to reroute with other assets in the field that could still pick those up in time and not cause further system disruption.
This kind of thing today normally takes quite a big concerted effort across many uh or many parts of the organization and many people looking at a lot of different records and a lot of different systems. The agent can do this all on demand from a single trigger and by the time anybody even knows something's happened they already have recommended action plans. In this case, if you guys are used to it, it's not a plan do check, it's a plan transfer repair.
But in this case again the agent generates the plan. Now the agent could be set up however you guys want. So it could be multiple plans with a with a priority or a singular plan. But the a plan is recommended. If it is chosen meaning if it is accepted then it moves to the transfer stage in this case which is okay we're going to take this out of service. We're going to get it into maintenance and we're going to transfer any necessary packages and then execute the repair.
Okay. As you'll note here, we have also continued to expand the awareness of the system. Not only was it how likely is this to fail, when will it fail, when do we have to reroute it so it doesn't fail on the street, but also the rerouting of the repair has already taken into account not only what is the failure mechanism, but where do we have the spare parts, where do we have the skills available within the repair service technicians to do this job, so that when this vehicle pulls in, we know already it can be done right there for the fast amount of time. and it has already provided not only a root cause but recommended actions and expected repair time.
So the system now can already be aware of when this asset is expected to come back into service for future agent uh decisions. Obviously the end of the day for this is we massively increase our opportunity to meet our our our agreements and ensure customer satisfaction with on-time deliveries or in this case also you know materials that didn't spoil or weren't perishable. However, in a more generalistic solution, this is the notion of how close to purely predictive maintenance can we get and always in operation supply chains or how close can we get to always on time deliveries by being able to understand even things like traffic patterns, dynamically rerouting around traffic, but also understanding time on road, other types of conditions, even down to have we properly accounted for loading the vehicle.
All of these things can be built in through the data acquisition uh through data bricks and into the agents for analysis. as I indicated, this isn't just about fleet ops, although we use this as an example. This is about anything that involves needing to use low latency data combined with organizational knowledge, whether it be through procedures, practices or other information even in other systems and bringing it together with an agentic solution to empower multiple organizations within or multiple functions within your organization to take better informed and more effective decisions and actions in near real time.
And with that, I'm going to leave this up here for a second because I wanted to talk again about one of the notions that really really impressed us in this in this project when we worked with CoNC was that between the ability to bring in vibration data at subsecond, I'm talking hertz type uh you know 1,000 10,000 kilhertz data coming into data bricks being processed and determined for anomaly at scale and time. and then being able to trigger an agent on demand for efficient and reliable operation that could then reply back to users in an interactive environment in minutes.
We did not expect as a whole we knew it would be we knew that Domo would allow us to do it quickly. We did not expect it would be less than 40 hours but everything together was less than 40 hours including the interface. So, uh, true unlock for your ability to use your your real-time data, uh, using data bricks and DOMA. And unfortunately, I give you guys five minutes back because in the panic to get this presentation up, I went quite quickly.
So, but I do appreciate everyone's time. Thank you.
Learn more about the Databricks Data and AI platform.
The information provided herein is for general informational purposes only and may not reflect the most current product capabilities or configurations.