AI Infrastructure

Operational AI for Buildings: From Data to Action | OP Ravi, Willow | TFiR

0

Buildings generate telemetry from thousands of sensors every minute, yet most facility teams still operate from static dashboards that describe problems only after they are visible. A stuck damper, an HVAC unit running at wrong setpoints in an empty room, or a battery cycling inefficiently against real-time grid prices: none of these are caught and corrected automatically. The gap between data collection and autonomous corrective action is costing building operators in energy waste, deferred maintenance, and missed sustainability targets.

In this interview on TFiR, OP Ravi, Head of Customer and Product Innovation at Willow, walks through how the Willow operational AI platform closes that gap, from ingesting live telemetry, documents, and 3D models through a unified knowledge graph, to deploying AI agents that detect faults, generate prioritized work orders, and send active control commands with an expert-in-the-loop review gate.

Guest: OP Ravi, Head of Customer and Product Innovation at Willow
Show: TFiR

Here is what every facility manager, platform engineer, and energy operations team needs to know.

Technical Deep Dive

Q: Why does building sensor data fail to produce actionable outcomes for facility teams?

OP Ravi, Head of Customer and Product Innovation at Willow, draws a clear line between data collection and value creation: raw telemetry sitting in a dashboard does not move a work order. Buildings can generate millions of data points per day from temperature sensors, fan speed readings, HVAC settings, access systems, and occupancy signals, reaching terabytes and petabytes at scale. Without a layer that normalizes those signals into relationships and triggers automated responses, the data describes the problem but does not fix it.

“Data alone does not create the value. Value is created when it helps people to make better decisions, solve real problems, and achieve some business outcomes or objectives.”

OP Ravi, Head of Customer and Product Innovation, Willow

Q: What are the three categories of building data that an operational AI platform must handle?

Ravi identifies three distinct data types that must be unified before AI can act on a building. Live data covers telemetry from sensors, actuators, and building systems including HVAC, elevators, and access management. Static data covers documents, architecture drawings, and sequences of operations. Spatial data covers 3D models and Revit files that encode geometry and physical relationships. Willow ingests all three through hundreds of connectors and normalizes them into a knowledge graph that forms the foundation of the operational AI platform.

“We have hundreds of connectors which can ingest the data from all over the systems, whether it’s BMS, whether it’s elevators, whether it’s IoT sensor, whether it’s different access management, your even Outlook calendar for building occupancy details.”

OP Ravi, Head of Customer and Product Innovation, Willow

Q: What does operational AI mean in a building context, and how is it different from analytics dashboards?

Ravi defines Willow’s operational AI as a combination of three AI paradigms working end to end in a single pipeline. Predictive AI uses historical data to forecast and simulate conditions. Perceptive AI reads raw sensor data and documents to detect anomalies and power copilot-style question answering. Generative AI produces recommendations, code, and agent-driven responses. A dashboard surfaces a reading; the operational AI platform ingests all three data categories, reasons across relationships in the knowledge graph, and either creates a work order or sends a direct control command.

“We are trying to create this platform which can use your all type of data and convert into some actionable insight and then even automate those insights based on the SME or expert in the loop.”

OP Ravi, Head of Customer and Product Innovation, Willow

Q: Is Willow’s operational AI platform specific to buildings or transferable to other industries?

The underlying architecture is domain-agnostic. Each industry has its own ontology defining entities and relationships: a power grid ontology maps generation through transmission and distribution to consumers, while a building ontology maps HVAC systems, floors, and spaces. Ravi notes his background applying similar digital twin concepts at Microsoft on IoT and at Google X on grid work. Willow is actively extending the platform to data center scenarios, treating data centers as a building type within the same mission of making every building respond to people, purpose, and environment.

“It’s transferable, but it needs some work. We are trying to build this agnostic platform that’s applicable to all type of industrial buildings.”

OP Ravi, Head of Customer and Product Innovation, Willow

Q: How does an AI agent workflow move from fault detection to a resolved work order in a real building?

Ravi walks through the end-to-end flow using Georgia Southern University, a Willow customer for approximately nine to ten months with hundreds of campus buildings onboarded. Live telemetry is normalized into the knowledge graph with full asset relationship context. The Activate technology runs 24/7, applying modular skills to detect anomalies, spikes, dips, or condition changes. When a skill fires, it generates an insight that can trigger a work order ticket for expert review, send a direct control command to the affected system, or dispatch an SMS or alert to the relevant team. The platform also surfaces the projected impact of taking the action, quantified as direct savings, cost avoidance, or risk prevention.

“Those insights basically can respond to now various ways: it can create a work order ticket, you can send a command back to the actual issue to resolve those issues, or it could send an SMS or alert or notify some group that there’s something happening.”

OP Ravi, Head of Customer and Product Innovation, Willow

Q: How does occupancy-based HVAC control work in practice as an AI agent use case?

Ravi uses occupancy-driven setpoint adjustment as a concrete example of active control. Traditional BMS systems run on static schedules and do not respond when a booked meeting room stays empty. The Willow platform combines calendar invite data, in-room occupancy sensors, and AV session signals to determine true occupancy state, then sends an active control command to adjust the HVAC setpoint in real time. The same pattern applies to battery systems: the platform reads real-time grid price signals and decides when to charge or discharge based on whether grid prices are high or low.

“By combining and normalizing all the data and deciding whether this room is really occupied or not occupied, and then sending the command, the active control command to the HVACs to set point or change the set point: that’s a real example that most of the customers are using.”

OP Ravi, Head of Customer and Product Innovation, Willow

Q: How does the system prioritize which building faults to fix first when problems outnumber available technicians?

Prioritization is customer-configured and dynamically adaptive, not a static rule set. Ravi explains that the platform accepts customer-defined goals such as cost savings, emissions reduction, ESG compliance, or risk prevention, and ranks faults accordingly. The same fault can have different priority depending on which floor it affects: a medical floor may outrank a sports facility under normal conditions, but a major event can invert that ranking in real time. The system evaluates priority across buildings and floors simultaneously, updating as conditions and occupancy data change.

“It’s not that you define a rule or you define the priority criteria and it’s done forever. It has to be adaptive also and it should be dynamic based on the conditions and the data.”

OP Ravi, Head of Customer and Product Innovation, Willow

Q: What does Northern Arizona University’s deployment reveal about operational AI at campus scale?

Northern Arizona University has been a Willow customer for multiple years across two campuses and has contributed significant product feedback. Ravi notes that early deployments required manual configuration: deciding building by building which operational AI skills to activate. AI-assisted onboarding now compresses that process from weeks to days or hours by using AI to classify and map the hundreds of thousands of BMS data points that previously required manual matching against documents and relationship schemas. NAU is now advancing into agentic HVAC control with occupancy-based setpoint commands, with an SME review gate that allows confident commands to progress to full automation without further human intervention.

“The same activity which used to take weeks, that same activity can be done in days or in hours today.”

OP Ravi, Head of Customer and Product Innovation, Willow

Q: How can building operational AI help solve the energy grid capacity problem driven by data center demand?

Ravi references IEA data placing buildings as the largest energy consumers on the grid, then outlines three flexibility levers: reducing overall consumption, shedding load during peak periods, and shifting load to off-peak windows. Because adding new grid generation capacity takes years through interconnection queues, demand flexibility from buildings can defer that investment. He cites a figure of more than 300 gigawatts of distributed energy resource capacity that buildings and DERs could contribute. Willow’s Active Efficiency module, called the Activate Pack, lets building operators participate in grid demand response programs, sending load shed and shift commands based on real-time grid signals.

“If we can optimize the building and that building’s load can basically either shift when there is congestion on the grid or can be reduced when peak demand is about to hit, those are the three different ways by which you can actually achieve a lot of demand flexibility.”

OP Ravi, Head of Customer and Product Innovation, Willow

Q: How do you prevent an AI agent from making a harmful mistake on physical building equipment?

Ravi is explicit that today the human retains full control. Every active control command produced by the Willow platform must pass through a customer-side expert review before it executes against live equipment. The progression to full automation is incremental: a command is approved for automated execution only after the customer has reviewed it, confirmed it does not cause issues, and verified it meets all business requirements. Operators can stop, pause, or resume any active control workflow at any time. Ravi frames the expert-in-the-loop design as a deliberate product choice, not a limitation.

“We have built this expert in the loop workflow rather than just agent go and do whatever they want.”

OP Ravi, Head of Customer and Product Innovation, Willow

Q: What data infrastructure does a facility team actually need before deploying operational AI agents?

Willow does not require customers to pre-install new hardware or add cost before onboarding. The starting point is whatever the building already has. A BMS system is the minimum viable input; a CMMS work order system adds further value but is not required, as Willow includes its own ticketing capability. 3D model files are ingested if available and skipped if not. The platform connects through existing connectors to whatever operational systems are generating data, normalizes that input into the knowledge graph, and begins running skills from there. Ravi summarizes the principle as meeting customers wherever they are.

“We never asked any of our customers that to start with Willow operational AI platform you need to first install XYZ or add more cost or effort on your site.”

OP Ravi, Head of Customer and Product Innovation, Willow

Resources & Documentation

  • Willow, operational AI platform for buildings: knowledge graph, Activate technology, Active Efficiency, and agentic control workflows
  • IEA (International Energy Agency), referenced for building sector energy consumption data and grid demand context

***

👇 Click to Read Full Raw Transcript

Swapnil Bhartiya: You may or may not know, but today buildings are generating more data than ever before. But most of it ends up on a dashboard that only describes the problem while a stuck damper stays stuck until a person goes find it and take a look at it. Willow has an answer for it. They call it operational AI agents that spot the issue, create work order and decide what to fix and when. Running today across more than 1,000 buildings at Northern Arizona University. And to talk about it today we have with us Op Ravi, head of customer and product innovation at Willow. Op, it’s great to have you on the show.

OP Ravi: Thank you Sapnil, it’s great to be here.

Swapnil Bhartiya: First of all, it’s my pleasure to talk to you. Of course we are going to talk about this challenge area but before that just to tell our viewers who we are talking to so that they also understand the gravity of the discussion. Talk a bit about yourself because you have great background and also talk a bit about Willow as well.

OP Ravi: I’m fortunate that in my career I have been turning data and AI and technology into some practical outcomes and customer solutions. I have been at Microsoft for around 19 years where I worked on IoT time series database or historian graph database with digital twin and lot of energy solutions. And after that I moved to Google X where I worked on the grid moonshot and worked on various innovative technologies and ideas which defining lot of energy transformation. And today as Soapnil mentioned, I lead customer engineering and product innovation at Willow focusing mainly on the customer outcomes and the value what we can deliver with the technology. And just very briefly what is how and what we are doing at Willow. At Willow I just want to start with our mission statement that hey, we want to create every building in the world on the planet actually respond to the people, purpose and the environment they serve. And as Sapnil mentioned that hey this buildings are generating tons of tons of data. Actually in my over these years actually one thing I have learned that data alone does not create the value. Value is created when it helps people to make better decisions, solve real problems and achieve some business outcomes or your objectives. And that’s exactly what operational AI at Willow and entire team is doing. So yeah, and we will do more discussion throughout this meeting. But yeah, that’s how what Willow is trying to do driving the data into some meaningful outcomes and the actions.

Swapnil Bhartiya: I talk to a lot of companies and they do work on these buildings because a lot of people may not even know that hey, when you talk about buildings, what data are you talking about? Because a lot of building safety, security, a lot of things. So if you can also just give them a context of when we are talking about data. Because people were like building data. I can understand cars, but buildings. But you know, there are a lot of you know, maintenance update. I mean security, a lot of features are there, you know, they have elevators and so just also give them a little bit of context so they do understand.

OP Ravi: At high level. You can divide the data into three main categories. There is a live data, which is your telemetry, which is coming from various sensors, various actuators, various systems, like even your temperature sensor or if your fan speed, if your HVAC settings and so on. The second category you can put like a static data which is mainly related to your documents and architecture drawings and sequence of operations and other artifacts. And the third category I would put like a special the 3D and geometry which is also your 3D modeling, Revit file architecture, drawing in the 3D models and those things. So we at Willow have hundreds of connectors which can ingest the data from all over the systems, whether it’s BMS, whether it’s elevators, whether it’s IoT sensor, whether it’s different access management, your even Outlook calendar for your building occupancy details. So we connect from via hundreds of connectors and then we basically ingest that data and normalize into our knowledge graph. And that knowledge graph is a very foundational core unit of our basically operational AI platform. So I hope that answers your question about the data. And when I’m saying data is like, it’s not tiny data, like if you imagine, we are actually having a lot of discussion on the efficiency of data also. And recently we were talking that, hey, if I have 100,000 data points and if I receive at a minute granularity or 15 minutes granularity, that’s going to create millions of data points in just a day and that’s going to be terabytes and petabytes of data. And so yeah, it’s a big data, huge data problem also. But it’s in the same time it’s an opportunity and it’s a gold mine which we help AI and operational AI help in that mining that gold.

Swapnil Bhartiya: And since you dropped the term operational AI at the very end, so that is also a very good latch and segue. Let’s start with the term itself. I mean, of course you call it operational AI as well. How do you define it? And how is it different from the analytics and dashboards that building teams have relied on for years and is operational AI different in this context, in the context of building versus when we talk in general as well.

OP Ravi: That’s a great question. So let’s start with some broader perspective on operational AI. So if you look at AI, we have often heard a term generative AI, digital twin, copilot, agentic approach, agents and what Willow’s operational AI platform means, it brings all those capabilities and technologies together and help our customers and organization make better decisions and automate the workflows. So as I said that hey, we are using AI even for connecting your various sources and onboarding your buildings, ingesting the data, normalize the data and then build into this operational AI platform. And if you look at again going back to same fundamental that hey, there are three, or I would say three main categories again from AI perspective also that hey, there is a predictive AI which basically use your existing or historical data to forecast something or predict something. That’s a predictive AI module. Then there’s a perceptive AI which is using your sensor data or your raw sensory data, whether it’s images or documents and detecting something from those. And then there is a third which is the most famous right now is the generative AI. So we have this LLM and agents and all those creating a response and code and other aspects come into this thing. When I’m saying Willow operational AI platform, we are using all three AI end to end in the entire pipeline or entire platform where we are using the predictive AI to forecast and simulate the data. We are using perceptive AI for reading your data and enabling copilot and answering any question reasoning and generative AI to have some recommendation creating the responses. Even our lot of engineers at the company using the code generation and we are also using agents for many aspects. So yeah, so that’s kind of operational AI in a nutshell. If I close with a statement that hey, we are trying to basically create this platform which can use your all type of data and convert into some actionable insight and then even automate those insights based on the SME or expert in the loop.

Swapnil Bhartiya: If we look at your operational AI definition or the platform technology you’re building, are they specific to Willows, you know, the whole building, you know, structures or this technology can be used somewhere else also.

OP Ravi: This technology can be used somewhere else also because as you as I introduced myself, I spent maybe almost a decade on the energy world. And if you look at the digital twin concept, each industry or each domain has an ontology on the modeling and then which is basically showing the entities relationship and all the graph. So in the grid you have the generation system that is passing to your transmission, distribution and the consumer. So you have the entire grid ontology and the industry. Same way we have the building ontology where you have how the HVAC system, buildings, labels, floors, land are connected and then on top of that you have all this AI capabilities which is generating those insights. So yes, it’s transferable, but of course it’s not that hey, it needs some work. If I need to apply the same operational AI technology in different domain like now we are heavily also working on the data center scenarios which you may have heard that everybody’s talking about the data center and energy. Now as our mission is again going back to that we want to create a world where every building on the planet can respond to people, purpose and the environment they serve. When I’m saying every buildings, it could be hospital, it could be airport, it could be data center, it could be an office, it could be a school, university. So yeah, so that’s why we are trying to build this agnostic platform that’s applicable to all type of industrial buildings and so on.

Swapnil Bhartiya: Now is it possible for you to walk us through a real example an AI agent spots stuck damper, what actually happens next from detection to the work order bring credit so it can actually be unstuck.

OP Ravi: Last month there was a NexusCon event I don’t know if you heard about in Detroit where our one of the customer, Georgia Southern University actually shared one story. So maybe I can double click on that one a little bit. So Georgia Southern is our customer for almost nine to ten months with us. And they have hundreds of building the entire campus has onboarded on that means. And when I’m saying the campus has onboarded that means there the live data coming to us. We already have the static documents and artifacts for reasoning and for answering any relationship and those details, operational procedures, questions. So if you walk through this process, so when you have the live data coming from those sensors or those things, those live data is going to be basically normalized into our knowledge graph associate with all the relationship whether hey, do we have any documents about that asset? Do we have any other relationship with that asset where that asset is located and so on. So that relationship is automatically built using normalization and those things. Then we have a technology called activate technology that activate technology basically runs 24 by 7. As you know, agents or AI never complains and they can run 24 by 7, they’re not tired anything. So we are leveraging fully so that 24 by 7 monitoring is happening via this our activate technology and that activate technology has lot of skills, that’s what we call skills. And based on those skills those modules or those skills will detect whether there’s any issue or any abnormal situation happen or this anomalies happen or there’s any other things happen or based on the trend there’s some new things happen which we never noticed. Like hey maybe some spike happen or some really a big dip happened. So on then those skills will generate insights and those insights and I’ll give you one real example, let me just complete the workflow. Those insights basically can respond to now various ways that hey, one is that hey it can create a work order ticket and then the subject matter expertise or the expert can review and act on those things. Another option is that hey, you can send a command back to the actual issue to resolve those issues. And other action is that it could send an SMS or alert or notify some group you pay there’s something happening. Maybe it’s not P0 or it’s not urgent, but someone need to look into that. And then once that action has been accomplished or done then the actual value will be realized. And we also show that hey if you take this action what will be the impact in terms of direct saving or cost avoidance or in other risk prevention those things. So this kind of a flow goes end to end. So very simple example could be that hey, which is true for almost all the customer that occupancy based HVAC control. I’ll give you one simple example and then we can maybe look at some complex example also. And when I’m saying occupancy it means that hey whether your room is occupied or not occupied. And traditionally BMS system has a monitoring capability but they don’t know that things has been changed. They have a same static setting which can stay on a schedule base or on season based or on some time based, right? But let’s say if you have a meeting scheduled from 9am to 11am but nobody showed up, so now what technically you should do. Maybe you should either dim the light, change the HVAC setting and reduce some temperature or set the point so that you can save some energy gain, efficiency gain, so on. So that’s one simple example. And then occupancy is also depends on many things, right? Hey, you can have the calendar invite, you can have the occupancy sensor in the room, you can use AV signal in the meetings. Hey, whether really someone actually logged in and started a meeting and so on. So by combining and normalizing all the data and deciding that hey, whether this room is really occupied or not occupied and then sending the command, the active control command to the HVACs to set point or change the set point on those HVAC. That’s a real example. That’s what most of the customers are using. But the same technology we have been using for let’s say the battery. Now you have a battery system. You can charge and discharge the battery based on different signal. Here different signal could be whether your energy price on the grid is at this moment is pretty high or low. And if it is pretty high, maybe I should use my battery and if it is low, let me charge my battery. So same workflow can work on various system and send different commands based on the skills and the actual system.

Swapnil Bhartiya: If you look at buildings, they have more problems than people available to fix them and things keep breaking. So how does the system decide what should be fixed first priority? Of course a lot of things which are like safety and all those things that do take priority. But in general how does the system decide what should be fixed first and when does the human need to step in or fully relying on AI?

OP Ravi: Yeah, no, so actually that’s a important aspect. So if you look at the initial challenge for all the building or facility managers or operators or know your world kind of thing that hey, they want to know what’s going on in my building and what systems and what is the state. So that was one initial challenge. Then the phase and the focus will move from know your world to like hey, activate and control your world. So that’s where the activate technology. And now with this modern era, we are looking from the operational AI perspective that hey, do we really need to babysit or do we really need to kind of monitor this 24 by 7. Otherwise if I need to do that job then I’m not going to gain and I don’t have a people to wash those system 24 by 7. So the prioritization plays a very important role. And it’s kind of a we need a customer input also that hey, what is the most important thing for you and based on that priorities input. Using the customer input. For example, a same system can save your dollar but maybe cause more emission. So now if customer goal is about sustainability then he would or they would prefer to maybe reduce the emission or achieve their ESG or sustainability goal rather than they’re saving the dollar. But now you can achieve both also together. Hey, you can look at hey, I want to optimize my action or prioritize based on the cost saving or the direct saving based on the future protection or risk, insurance or prevention. So there are multiple dimension or the input has been considered to decide the priority and when let’s say now you have two system with the same settings. Now you need to look at even the building by building. Also that hey, which building I should pick or which floor I should pick compared to. So one floor could have let’s say medical center and another floor could have, let’s say the sports facility. Maybe I should prioritize accordingly based on the customer input whether sports facility has a more importance priority compared to medical department. Of course medical may have more importance, but let’s say if there’s some important event going on, then all of a sudden maybe the sports facility also become an important from that perspective. So yeah, I think that’s dynamicness or that flexibility or adaptability come based on the data is very important. It’s not that you define a rule or you define the priority criteria and it’s done for the forever. It has to be adaptive also and it should be dynamic based on the conditions and the data.

Swapnil Bhartiya: You folks, you know of course Northern Arizona University is running this across a real campus at a scale. What does that deployment tell you folks about how operational AI performs in live building environments?

OP Ravi: NAU is one of our flagship customers. Like hey, they have been using Willow for many years now and they are providing a lot of great feedback into our product also. So they have two campuses, south campus and north campus. And there was a very interesting actually I run even the capstone project with the students in NAU which is also very interesting story to cover some another day. But in short, see this is a journey which we are going through with all our customers. And initially we did a lot of deployment manually. Like hey, we are doing with customers that hey, which campus is more important, which building is more important, where you think, which operational AI will help you in which buildings or which things and so on. So initial deployment was a lot of things were manual but they give a lot of great feedback, hey, why not about this thing or that thing. And now we are using the operational AI or AI onboarding also as part of this with the new customers. So now let’s say the same activity which used to take weeks, that same activity can be done like in days or in hours today. So for example, when we onboard a building, each BMS system has hundreds of thousands of points which we need to classify, which we need to basically discover and classify. And that classification now can be done by AI. And that AI onboarding can help us to even save those time. Because if I need to manually look at those hundreds of thousands points match with the documents or the relationship. Now the AI is helping us in this NAU case also. And now NAU is kind of advancing into more into this agentic control things that hey, now we can set their HVAC settings based on occupancy, based on other conditions and can send the command, automatic command also. So there are two. One point I want to highlight here is that when we say automatic command or active control, the future is fully automation. But not everybody is there today because there is a review or the expert review or some kind of a check and balance need to be maintained before we can say that hey, it can go without any review. So we have a mechanism where an SME from customer side can review and approve that, hey, yes, I have a full confidence on this active control command. It can go full end to end without any human intervention. So that’s another very good functionality which we have for our customers.

Swapnil Bhartiya: Now let’s look at different aspect and that is a huge discussion these days, which is like AI is pushing electricity demand almost everywhere. A lot of discussions are going on. A lot of FUD is also going against data centers as well. Now it depends on how you look at it. Whether Almel farms consumes more water or data center consumes more water, that is out of discussion. But how can operational AI help buildings become more efficient with the energy that they consume? What role can it play to solve that problem as well?

OP Ravi: I would say that’s the huge benefit which we can leverage from buildings. Let me take a step back. So when I say again, going back to our mission statement, right, we want to build this, create a world where every building can respond to the people and purpose and the environment. And a building has lot of energy consumption happening. Because if you look at the latest report from IEA, the International Energy Agency, then building domain has consumed the most of the energy from the grid. Now there is an energy efficiency which can divide into again into multiple ways to achieve that. One is the simple, you can reduce the overall consumption. The another one could be that hey, you know that hey, these are your peak hours so you can shed the load and the another one is the shift the load. So if you combine those three functionality, the energy efficiency overall, the shedding the load or shifting the load is the best way. So now why and how buildings can help you in this flexibility. So now assume you are having a grid and grid has a base power as well as the peak load. And most of the grid has to be built to serve their peak load. Otherwise there will be an outage and other scenarios. Now based on this data center demand, now the grid need to increase their peak load and add more generation. Now to add a generation plant or more peak load support. It’s a multi, multi year process because there’s an interconnection queue that queues need to be fulfilled, that criteria need to be fully met. Then only the grid can approve to add a new generation plant or so on. But if I have the buildings where I can manage the load and reduce or contribute back to the grid as a demand flexibility, that’s where the buildings can play a big role. And that’s based on if I remember correctly, there was around 326 or some 300 more than 300 gigawatt of distributed energy resource capacity was can be served by buildings and the DERs in the grid. So if we can optimize the building and that building’s load can basically either shift when there is a need on the or the congestion on the grid or can be reduced by when there is a peak demand is about to hit. Those are the three different ways by which you can actually achieve lot of demand flexibility. And that way you don’t need to build actually you can even defer the entire development of the new generation plants. Or you can basically delay that and achieve the same goal with the existing capacity using building flexibility. I’m very much fortunate and this grid interactive efficient building work which we are doing in the willow, we call that active efficiency as a module or as activate pack where we allow our customer to participate in the grid programs and help them to load shed and shift the load based on the grid signal.

Swapnil Bhartiya: Earlier we were talking about autonomous, fully autonomous. One of the challenge, first of all, we cannot trust AI fully. But what happens when an agent gets it wrong, makes a mistake in a physical system you cannot just roll back like software. So how do you earn the trust of facilities before letting AI act on their equipment at the same time, how much decision making power humans have there?

OP Ravi: Today they have a full power. So a human has a full power to control stop, resume any active control commands which we have in Willow platform we allow, as I said, that the active control workflow is based on the skills and the conditions and the rules which we have defined with help of customers and their inputs and their thresholds the flow. Today what we have is that hey, initially customer side need to review before that command can change the actual setting or change the actual work. Once the customer feel comfortable and they see that that thing is not causing any issue or that is meeting all the business requirement then they can approve that same command for automated workflow. So that automated workflow is not the first step in that agent workflow today. The I think that expert review is required I feel at least in the beginning until unless we have this full fledged autonomous or fully automated services agent workflow or those things expert can have a full control today. That’s what I can say and we are very proud of that part that hey we have built this expert in the loop workflow rather than just agent go and do whatever they want.

Swapnil Bhartiya: For an operational teams. Who’s watching this? Where do they actually start, what data has to be in place, how they prepare itself before the agent can act on their building and how deeply do you work with them? Is it more like turning solution they come in or you actually do a lot of whole hand holding and actually work with their teams as well.

OP Ravi: We meet wherever they are. We don’t want to add additional workload or even cost to our customer. We never asked any of our customers that hey to start with Willow operational AI platform you need to first install XYZ or add more cost or an effort on your site. So that’s very much our step one that hey we want to meet customer wherever they are. So if customer has a let’s say 3D modeling file we will ingest and give you the 3D digital twin. If they don’t have 3D file that’s not a problem at all. We will do whatever you have. So if you have a BMS system which is the core which I think every building has that so we’ll start with BMS system. If you have a work order CMMS system we will ingest your integrate with that. I think with BMS and CMMS those two systems should allow us to achieve that. Even many of our customers are also using Willow’s ticketing tool also that hey you may not need a work order management system also. So as long as you have the BMS and various systems of the where your data is generating or whichever the system you want to monitor we will start with there.

Swapnil Bhartiya: Op Ravi, thank you so much for walking us through how operational AI actually moves from describing buildings to actually acting on them and making things automobiles. Also explain that people may not even realize that the buildings that they live in some of those how much software is actually managing all those systems. So thank you so much. And of course, those who do want to learn more about it, please go and check out below in. And Opie, once again, thank you for your time and I would love to have you back on the show.

OP Ravi: Thank you, and great to be here. Thanks for having me.

Kubernetes v1.37 Deprecations to Act On Now | Dipesh Rawat, CNCF | TFiR

Previous article

AI Inference: Cut Costs Without Losing Control | Jon Alexander, Akamai | TFiR

Next article