Venkat Kondepati has spent more than two decades building data platforms across financial services, energy, and utilities. His career started in GIS, digitizing stormwater pipelines and building spatial databases. This foundation fundamentally shaped how he approaches enterprise architecture today, where he argues that AI models perform better when they understand where things happen.
We spoke with Kondepati about why enterprise leaders overlook location data, the technical barriers to putting spatial AI into production, and how generative AI will reshape industries when combined with geolocation.
We asked Kondepati what pulled him toward combining spatial intelligence with AI, and he was quite direct about the stakes.
“Any AI solution without location information can provide a generic solution and miss a meaningful context. If we are talking about energy prices, there are significant differences between the APAC region and NA. Buying habits differ between developed and developing countries. In India, a car is a luxury for many, but in developed countries, it is essential. Enterprise leaders who don’t consider geolocation when making important decisions or when trusting data and AI recommendations can go wrong if they don’t believe in ‘the science of where.'”
He points to the practical applications he has worked on over the years: land and parcel management to support property tax planning; watershed development; site selection for Pizza Hut locations; and load analysis for electric companies serving hospitals and emergency facilities. He has mapped acreage-grading solutions for geologists and optimized supply chain routes using map functionality.
“When we include location information, the traffic congestion changes, and supply chain constraints differ from port to port. The agricultural industry impacts the entire production and economic conditions. The infrastructure, water, gas, sewer networks, and transportation industries feel disconnected. Including the spatial intelligence variable in AI models provides additional context and improves predictive accuracy.”
“Enterprise leaders view spatial intelligence as optional, but it is not, given the significance of geopolitical conditions.”
GIS and AI typically live in separate technology stacks. Kondepati explains why that separation has persisted and how cloud infrastructure is changing the equation.
“Traditionally, GIS experts and tech stacks are isolated from the rest of the world and run their workloads on desktops or dedicated servers to maintain their own workflows due to the complexity and differences in working style. Even the GIS software is complex, with geometric integrations, and maintaining data accuracy and integrity is a challenge.”
“But with cloud and AI, we can store data with lat, long information on a distributed platform like Snowflake, Databricks, or Apache Spark on YARN, allowing data scientists to access the data when needed without any data transformations. We can easily perform spatial operations on a large dataset using cloud datastores to identify interrelationships. GeoPandas is one of the best examples of using Spark and GIS to perform operations or integrating Apache Sedona on top of Snowflake or Databricks infrastructure, to find houses within a 3-mile radius of a given restaurant.”
He describes a project using location data to predict electricity prices from historical data and forecast prices over the next 30 to 50 years, enabling large-scale industries to plan their operational expenses.
“We used ML models and location information together to identify alternative substations when the price at one location is higher.”
When enterprises try to put spatial AI into production, they hit specific technical problems. Kondepati has solved them through parallelization and cloud orchestration.
“In my experience, 3D modelling is one of the cool features of GIS, and applying it to large datasets and performing analysis is challenging. I solved this problem by slicing the data by US county, running the models in parallel, and orchestrating them using a cloud service, which helped expedite processing time. I have reduced the total duration from 30 hours to 90 minutes using AWS Fargate, ECS, an S3 bucket, and Step Functions.”
“The second problem is that most GIS data is stored in ESRI ArcGIS format and needs to be exported and made compatible with AI tools in GeoJSON or GeoParquet format. Modern big data platforms do not support storing complex geometries and often fail to do so. For example, multi-part polygons or multi-point features are challenging, and performing spatial operations either fails or produces incorrect results.”
Kondepati has managed distributed engineering teams across continents for more than 15 years. His approach to driving AI adoption comes down to giving people time to learn.
“My simple answer is to collaborate asynchronously on Teams and share the knowledge. We started with GitHub Copilot knowledge-sharing sessions and Power Automate workflow automations, later developed our own version of ChatGPT, and had many lunch-and-learn sessions to encourage collaboration.”
“I always believe in investing in people by allowing them to spend 10% of their time each week learning or experimenting with something new. My team members used to block time to learn or build quick prototypes. This helped us a lot in keeping the team up to date with the latest technology and adapting to change. We also conduct group projects, exercises, and activities. I have personally undertaken serverless training on AWS Cloud as a Cloud professor, with practical, step-by-step instructions.”
AIOps and GenAI are changing how enterprises run infrastructure and data pipelines. Kondepati sees spatial intelligence fitting into that shift, but warns against keeping capabilities separate.
“AIOps and GenAI are changing the way enterprises manage their cloud deployments, improve observability, and run SRE functions. Some organizations are taking advantage of modern infrastructure patterns and adapting to change, while others are struggling to do so.”
“I have implemented a simple strategy to explore and adjust to the new technology to gain insights, but it has had no impact on business operations. If we are keeping the changes in Agile and an iterative approach, we can’t manage them by following a big-bang implementation strategy. We also have third-party GIS software compatibility issues with some critical functionality, so I have separated them and moved 95% of the existing functionality into serverless architectures using containers. GIS software needs specialized operating systems for specific use cases. If we keep these separate, it is hard to cope with change and meet customer demands in the long run.”
When talking to enterprise leaders about applying AI to location data, Kondepati finds the same misconception repeatedly.
“Enterprise leaders think that geospatial data means adding extra columns to store the geometry as a GeoJSON object and using it when needed; otherwise, 95% of use cases are attempted without spatial relationships or context. I have observed this many times and need to clarify the importance of geospatial analytics and how they solve real-time issues efficiently.”
He offers examples: an energy company evaluates the impact of gas emissions on the environment and health hazards using spatial overlay and pollution maps. Supply chain disruptions can be managed through route optimization and fleet management during natural disasters. E-commerce companies are using drones to deliver products to homes. “In any case, geospatial analytics allow us to identify the trends and patterns to assess the alternative solutions.”
Looking ahead, Kondepati sees spatial intelligence, cloud-native architecture, and generative AI combining to reshape energy, financial services, and supply chain operations.
“The current manual surveys will be replaced with drones, get live data fed to Gen AI workflow automations, and make decisions on the fly with cloud-native architectures. The complex models leverage location intelligence and adapt to make decisions using multiple variables while following regulations.”
“The use of data across industries has become the new norm, and the intersection of agentic AI, Gen AI, and live geolocation enables immediate decision-making with fewer human interventions. We can detect forest fires and measure harvesting, predict soil erosion, and even improve utility infrastructure.”
“The increased demand for electric energy from EVs will put tremendous pressure on the system and require alternative solutions. The combination of spatial intelligence, Gen AI, agentic AI, and cloud-native solutions will support the change and effectively assess its impact on our environment.”
“The live location with quality images will help to make faster decisions. The image recognition services will use geolocation and coordinate-based tagging to provide accurate information.”
