Drumil Joshi Unveils AI Blueprint for Grid Resilience at PVPMC 2025

Interviewer: Drumil, first off, congratulations on presenting the poster at the prestigious PVPMC 2025. What was the core focus of your presentation?

Drumil Joshi: Thank you! At PVPMC 2025, we presented a real-time, data-driven analytics solution that transforms raw BESS telemetry into instantly actionable insights: we ingest and cleanse high-frequency data from power set-points to feeder measurements, compute a single Oscillation Severity Index that quantifies the magnitude, duration, and amplitude of any fluctuations, and automatically flag and alert on emerging anomalies all visualized through an intuitive Dash dashboard with live gauges, trend charts, and contextual weather overlays so both engineers and decision-makers can spot issues early, prevent costly downtime, and optimize the life and performance of energy storage assets.

Interviewer: That sounds highly advanced. Could you elaborate on what makes your system truly novel?

Drumil Joshi: What sets our system apart is the way we turn mountains of high-frequency BESS data into a single, intuitive health metric the Oscillation Severity Index by dynamically weighting the size, duration, and frequency of each fluctuation, rather than simply logging alarms after the fact. We've paired that with live weather context so operators instantly see how temperature or grid conditions might be influencing behavior, and we've wrapped it all in an interactive Dash dashboard complete with real-time alerts, automated reporting, and even a natural-language "SPC Helper" chat assistant for on-the-spot questions. This isn't just another monitoring tool; it's a proactive intelligence layer that predicts and prevents issues before they snowball, giving both engineers and executives the confidence to keep energy storage assets running longer, safer, and more profitably.

Interviewer: That's impressive. What specific challenges were you solving when developing this system?

Drumil Joshi: When we built this system, our biggest hurdle was wrestling with the sheer volume and messiness of high-frequency battery data spikes, dropouts, and inconsistent timestamps made it all too easy to drown in noise or miss a critical deviation. We needed a way to distill thousands of data points per second into a single, reliable health metric without triggering endless false alarms, so we developed the Oscillation Severity Index with dynamic weighting to focus on truly impactful events. At the same time, we had to marry that score with real-time weather and grid context, so operators understand "why" something is happening, not just "when" it's happening. Finally, delivering these insights in an intuitive, interactive Dash dashboard complete with automated reports and a chat helper meant solving performance and usability challenges so that engineers and executives alike could act on the data immediately and with confidence.

Interviewer: Let's talk about features. What can an operator do using your AI-based dashboard?

Drumil Joshi: With our AI-powered dashboard, an operator can instantly see the health of every BESS asset through a live Oscillation Severity Index gauge, drill down into time-series charts to pinpoint exactly when and where a fluctuation occurred, and view side-by-side weather or grid-condition overlays to understand external drivers. One click brings up automated incident reports summarizing key metrics over any time window, while interactive filters let you compare performance across sites, battery strings, or operating modes. If you need deeper insight, the built-in "SPC Helper" chat lets you ask plain-English questions like "How many oscillation events did Site A have yesterday?" and get immediate answers without hunting through raw logs. In short, the dashboard turns complex data into simple actions: spot a looming issue, investigate its cause, and export a compliance-ready report or alert your team in seconds.

Interviewer: With so many assets, how scalable is your system?

Drumil Joshi: Our platform was built from the ground up to grow with you: it ingests thousands of streaming data feeds in parallel, so adding a new site or ten simply means spinning up another data-collector service no reconfiguration required. Behind the scenes, our preprocessing and OSI calculations run across a cluster of worker nodes that automatically scale to match your data volume, while the Dash dashboard is containerized so you can deploy multiple instances behind a load balancer. Whether you're monitoring five sites or fifty, the system dynamically allocates compute resources to keep latency low and performance snappy, and operators see a unified view of every asset without any slow-downs or manual tuning.

Interviewer: Can you tell us more about the technology stack behind the system?

Drumil Joshi: Under the hood, our solution is a pure Python stack orchestrated for real-time scale: we stream operational data into a time-series database (we've standardized on InfluxDB), run preprocessing and our custom Oscillation Severity Index calculations in Python workers managed by Apache Airflow, and push results through an API built on FastAPI. On the frontend, we chose Plotly Dash for its seamless integration with our Python analytics everything from live gauges to trend charts is rendered in the same language that crunches the numbers. Containerization through Docker and Kubernetes lets us spin up new data collectors or dashboard instances on demand, and we tie it all together with Redis as a lightweight message broker so alerts and chat queries flow instantly between services. For contextual weather data, we tap a public API and join those feeds in our pipeline before visualization. The result is an end-to-end, cloud-native architecture that's simple to extend, easy to maintain, and lightning fast for operators and executives alike.

Interviewer: This system seems like a game-changer. Have you had interest from other companies or collaborators?

Drumil Joshi: Absolutely even before we wrapped up development, several groups reached out to explore pilots and partnerships. NREL colleagues were keen to integrate our OSI metrics into their broader PV+storage modeling workflows, while Sandia's PVPMC community invited us to demo at their next technical workshop. We've had conversations with major grid operators who see value in marrying our analytics with their SCADA platforms, and a couple of energy storage OEMs are evaluating how to embed OSI directly into their control firmware. On the industry side, consulting firms and system integrators are lining up to offer the dashboard as part of turnkey monitoring solutions. In short, the response has been overwhelmingly positive, and we're already working on joint trials that will expand the system's reach well beyond our initial sites.

Interviewer: Finally, for young engineers or tech enthusiasts looking to break into the AI + energy space, what's your advice?

Drumil Joshi: My advice is to blend a solid grounding in power systems with hands-on AI experimentation: start by mastering Python and the core machine-learning libraries (scikit-learn, TensorFlow or PyTorch), then apply them to real energy-domain data whether that's public grid datasets or your own small-scale solar or battery logs to build simple models that predict voltage swings, state-of-charge, or equipment anomalies. Pair that with foundational coursework or self-study in electrical engineering so you truly understand how energy flows, where inefficiencies arise, and why a well-tuned algorithm can make or break grid reliability. Seek out open communities like the PVPMC forums or IEEE working groups, contribute to GitHub projects, and don't be afraid to reach out to mentors at national labs or utilities for informal advice. Finally, cultivate a storyteller's mindset practice translating your technical findings into plain language and impactful visuals (dashboards, infographics, short videos) so you can bring both engineers and executives along on your AI-powered energy journey. That combination of deep technical chops, domain fluency, collaborative curiosity, and communication skill will set you apart in this rapidly evolving field.

Conclusion:
As an editor, I'm struck by how Drumil Joshi's AI-driven framework transforms what was once reactive troubleshooting into proactive system stewardship: by uniting high-frequency BESS telemetry with a purpose-built Oscillation Severity Index, he gives operators a clear, quantitative "health check" for each asset and a contextual map that links performance dips to weather or grid conditions. This approach doesn't just catch problems it anticipates them, guiding maintenance before small anomalies ripple into costly outages. Beyond its technical elegance, the platform's intuitive dashboard, automated reporting, and conversational "SPC Helper" chat demonstrate a rare fusion of data science rigor and user-centric design. In doing so, Joshi's work not only enhances the operational resilience of today's energy storage fleets but also sets a new standard for how AI can drive cleaner, more reliable grids worldwide.