#business & career solutions | Explore Tumblr Posts and Blogs

jcmarchi · 2 months

Text

Vikhyat Chaudhry, CTO, COO & Co-Founder of Buzz Solutions – Interview Series

New Post has been published on https://thedigitalinsider.com/vikhyat-chaudhry-cto-coo-co-founder-of-buzz-solutions-interview-series/

Vikhyat Chaudhry, CTO, COO & Co-Founder of Buzz Solutions – Interview Series

Vikhyat Chaudhry is the CTO, COO and co-founder of Buzz Solutions and a former data scientist at Cisco, a machine learning/embedded systems engineer at Altitude and a Stanford graduate.

Buzz Solutions delivers accurate AI and predictive analytics software to power more efficient visual inspections for transmission, distribution, and substation infrastructure.

Can you share your journey and career highlights that led you to Co-Found Buzz Solutions?

I grew up in New Delhi, India, with a natural curiosity for innovation and engineering and I attended the Delhi College of Engineering where I studied Civil and Environmental Engineering. I particularly remember a moment during my final year when I built a drone from scratch and flew it in the city. The assignment was to monitor air pollution in New Delhi and through this experiment, I found that the quality was above 500 AQI, which is the equivalent of smoking 60 cigarettes a day. The poor air quality could be directly traced to a lack of electrification, rising vehicular emissions and increased number of coal-powered power plants over the years. This experience solidified my interest in using technology to address real-world problems associated with energy and power.

Before founding Buzz, my technology background led me to my role as the Lead of Machine AI and Data Science Teams at Cisco Systems for a few years. This experience was invaluable and built my exposure to a diverse range of artificial intelligence and machine learning projects early on.

I received my masters in Civil/Environmental Engineering from Stanford University in 2016. During this time I took classes specializing in energy engineering, building my interest that started overseas. I met my co-founder Kaitlyn in a class where we bonded over our passions for the environment, energy and entrepreneurship. We stumbled upon a great need in the utility industry and have been working on solutions to address it ever since.

What key developments have you observed in the progression from traditional AI to Generative AI during your career, and what significant impacts has this transition had on various industries?

In 2022, we began experimenting with Generative AI. GenAI in the utility sector is an interesting use case because the data we work with involves many different variables. There are factors like camera resolution, angle of capture, and object distance – and those are just for the drones. There are also environmental conditions like corrosion or vegetation encroachment that introduce numerous degrees of freedom. Because of this complexity, good training data for grid models can be hard to come by.

That’s where GenAI has come in over the past few years – as artificial intelligence and machine learning improve, so do the training sets it creates.

GenAI has become a viable option for training models, especially with crucial ‘edge cases’ where variables have more extreme values, such as in the case of a wildfire. As GenAI in the utility industry progresses, synthetic data sets, based on real world data, will help in further training models to handle complex and unique data scenarios more effectively, offering significant improvements in predictive maintenance and anomaly detection which will in turn reduce natural disasters.

Can you elaborate on how Buzz Solutions’ AI tool uses real data for anomaly detection and the benefits it offers over synthetic data?

In the utility industry, real data means whatever can be captured in the field, usually including images or video taken from aerial sources like drones or helicopters. Synthetic data, on the other hand, is data collected through an image replication process that manually alters various components of an image to try and account for an exponential amount of scenarios and edge cases. Currently, it’s great on paper but not in practice. Models trained with real data from the start are proven to be more accurate and the advantage is that through the use of real data, teams can map 1:1 with the ‘ground truth’ – an accurate representation of the physical world scenarios a technician is likely to encounter (like background noise and weather). The real data accounts for real-world possibilities, and includes the unpredictable variables of fault detection.

While synthetic data alone is not able to optimize for real-world scenarios (yet), it still plays an important role in training models.

What are the biggest challenges you face when integrating AI with legacy systems in utility companies?

Legacy systems in utility companies are often incompatible with AI advancements. Two major challenges we see companies face are internal transformation and data management. Siloed data and communication can be detrimental to digital transformation efforts. The data that utilities already possess must be managed and secure while information is carried over.

Additionally, utilities that still use on-premises data storage face larger challenges. The shift from on-premises data storage to cloud infrastructure is not the issue, but rather the extensive transformation and aftershock that follows. This process demands substantial resources and time, making it difficult to add different technologies on top of the transition. Introducing effective AI solutions is not recommended until this process is complete.

It’s also important that internally, there is a cultural shift along with the technology shift. This requires having employees on board with continuous learning and adaptability to changes in the process and looking at AI solutions as effective tools to make their day-to-day jobs easier and efficient.

Can you explain the process of training AI models with field-tested data from vital infrastructure sites?

A huge part of the training process is ingesting the aerial data provided by drones and helicopters. We choose to use drones over methods like satellites due to the flexibility and immediate data delivery that they allow. We use three main different types of algorithms: image clustering, segmentation, and anomaly detection.

Our technology is driven by Human-in-the-loop machine learning – which allows subject matter experts on our team to give direct feedback to the model for predictions below a certain level of confidence. We are lucky to have the SMEs on our teams that we do – with their decades of combined field technician experience, they provide feedback to make our models more accurate, personalized, and robust.

By using real field-tested data, we can ensure that our anomaly detection is highly accurate and reliable, providing utility companies with actionable insights.

How does Buzz Solutions’ AI technology contribute to making power line repairs safer?

Power line repair work is one of the deadliest occupations in America, and the industry is experiencing the effects of an aging workforce and technician shortages.

With our technology, PowerAI, emergency response has been made more effective and accurate, so that technicians can assess damage remotely and have time to develop a predetermined course of action – which reduces the possibility of sending in a technician to an unknown, potentially dangerous situation.

PowerAI uses computer vision and machine learning to automate a huge portion of the fault detection process. It has made the analysis of large masses of data points faster, safer, and cheaper, so now the technicians face reduced unnecessary risk and higher operational efficiency. This operational efficiency presents itself through smaller costs, quicker turnaround times, and preventative maintenance.

What role do drones and other advanced technologies play in modernizing infrastructure inspections?

Historically, the process of infrastructure inspections was completely manual and very mundane. Inspectors would sit in front of the computer screen, shuffle through thousands of images, and identify issues by hand. This process became unsustainable when power lines kept experiencing issues leading to more unsafe situations and higher regulatory overviews, increasing the amount of data needed to be reviewed in a shorter amount of time.

AI-based technology significantly streamlines the process of analyzing data, which reduces the time and cost involved. This allows utility companies to deploy repair teams more quickly and effectively. The detection of issues is also a lot more precise, ensuring that repairs are timely and preventing burgeoning hazards.

In capturing images for analysis, drone inspections are safer and more cost-effective than other methods of infrastructure like helicopters, satellites, and fixed-wing aircrafts. Their portability allows them to maneuver in a way that they can get close and capture more granular information.

How does Buzz Solutions’ AI-powered platform help utility companies with predictive maintenance and cost savings?

Our solution takes most of the manual analysis work out of grid inspection. PowerAI can quickly identify dangerous situations to prevent potential disasters and provide critical information for monitoring and security purposes. The AI algorithms are trained to identify anomalies like extreme temperatures, unauthorized vehicle access/personnel, thermal imaging, and more.

On top of preventive tracking, PowerAI can also provide tiered prioritization of anomalies for optimized maintenance planning. All of these things minimize the need for physical inspections, reducing operational costs and safety risks associated with manual inspections. The AI-powered platform also provides more precise and accurate detection, improving maintenance decisions.

Can you discuss the impact of adopting AI on the operational efficiency of utility companies?

After the initial lift of adopting an AI model, a utility company will continue to reap the benefits of the model for an endless amount of time. The lifecycle of an AI model begins at installation. AI can harvest actionable insights from thousands of images taken across hundreds of miles of infrastructure. Considering that we received our first dataset from a utility on a tape, this is extraordinary and it’s only getting smarter. AI makes early detection of maintenance issues much more possible, which prevents minor incidents from escalating into larger safety hazards like wildfires and serious injuries. It reduces the need for human inspections, making the utility more cost-effective.

In your article “Adopting AI Is Just The Beginning For Utility Companies,” you discuss the initial steps of AI adoption. What are the most critical considerations for utilities starting their AI journey?

There is a huge opportunity for utilities to use AI, and many solutions out there to consider. Before jumping in, it’s important to identify your goals and set a stable foundation – what challenges are you currently facing that you would like AI to help address? Does your team possess the technical expertise and time to take on such a complex overhaul? How will it impact your customers?

On top of being aligned internally is being prepared to get more data than the utility has previously, which will likely lead to more maintenance as issues arise. A utility should have a plan to accommodate these requests and be sure that they have the proper resources before starting their AI journey. Utilities also need to work with solution providers to implement the right data access, privacy and security when deploying AI solutions. AI-generated insights should finally be fed into existing utility workflows so that they become actionable and can meet the business and operational goals of the organization.

Thank you for the great interview, readers who wish to learn more should visit Buzz Solutions.

0 notes