AI Meets Energy: How Drumil Joshi is Steering a $450M Renewable Fleet with Predictive Intelligence

Introduction:

At the intersection of artificial intelligence and clean energy stands a quiet revolution led by innovators like Drumil Joshi. At PVPMC 2025, one of the most influential platforms for photovoltaic and energy system modeling, Drumil presented a solution that might just redefine how we manage Battery Energy Storage Systems (BESS). In this exclusive interview, he walks us through his pioneering work, what makes it game-changing, and how one person backed by AI can manage a renewable portfolio worth $450 million.

Interviewer: Drumil, your presentation at PVPMC 2025 gained a lot of attention. What was the core innovation you introduced?

Drumil Joshi: At PVPMC 2025 I introduced "PI Data-Driven Predictive Analytics for BESS Performance Monitoring," a poster that condenses the entire life-cycle of Battery Energy Storage oversight into one streamlined loop: high-resolution PI historian feeds are ingested, denoised, and millisecond-aligned; the cleaned stream drives our physics-aware Oscillation Severity Index (OSI), a weighted blend of magnitude, duration, and amplitude that instantly tells operators whether a pack is healthy or veering toward instability. Because persistent oscillations can slash efficiency by up to 30 % and drain hundreds of thousands of dollars per site each year, early detection is business critical. Our anomaly-detection module patrols the OSI in real time, tagging "oscillation-prone timestamps" and routing them to an interactive Dash dashboard that couples gauge visualizations, trend graphs, and a "Chat with SPC Helper" letting engineers query the data in plain language and receive instant, context-rich answers. Additional layers such as real-time climate overlays that correlate weather swings with battery behaviour and automated report generation for compliance transform raw telemetry into actionable insight without human babysitting. By uniting self-healing ETL, an explainable risk metric, and operator-centric visualization, the poster demonstrates a practical, vendor-agnostic blueprint for cutting downtime, extending asset life, and safeguarding grid reliability proof that thoughtful AI can turn a flood of data into a single, decisive heartbeat for a modern BESS fleet.

Interviewer: That sounds very advanced. What prompted you to build this?

Drumil Joshi: A vivid public warning came from the joint NERC-WECC report on the March 9 and April 6 2022 California battery-storage disturbances: two normally cleared faults rippled through several utility-scale BESS sites, tripping inverters in masse and exposing how undamped control oscillations can slash output and delay grid recovery essentially the very pathology the OSI now flags in real time. nerc.com Equally telling is EPRI's May 2024 study of its Battery Energy Storage System Failure Incident Database, which dissected 81 documented BESS events and found that more than a third of all downtime was rooted in control-system oscillations and protection mis-trips problems the authors conclude could have been averted with continuous performance analytics and early-warning metrics like ours.

Interviewer: And you manage all of this across a massive energy portfolio?

Drumil Joshi: Absolutely. I oversee 33 utility-scale solar plants, 15 wind farms, and our two-flagship battery-storage sites roughly $450 million in operating assets because every critical workflow is automated to shrink the cognitive burden. I single-handedly manage diagnostics analytics across the fleet using this AI platform. It enables predictive diagnostics, automated reporting, and contextual weather-based insights.

Interviewer: Let's talk about features. What exactly can your system do?

Drumil Joshi: The poster showcases a system that goes well beyond a single index: after ingesting and cleaning high-resolution operational feeds, it pinpoints oscillation-prone timestamps, overlays them with real-time climate data to reveal weather-driven vulnerabilities, and automatically compiles compliance-ready reports that land in operators' inboxes without manual effort; on the front end, an interactive Dash dashboard combines an OSI gauge, detailed performance graphs, and long-term trend analyses with a natural-language "Chat With SPC Helper," letting engineers ask ad-hoc questions and receive instant, data-driven answers all of which accelerates troubleshooting, sharpens forecasting, and keeps Battery Energy Storage assets in the green.

Interviewer: What was the response like at PVPMC 2025?

Drumil Joshi: The reception was electric: within minutes of the poster session opening, the aisle around our display was four-deep with utility engineers, battery OEMs, and academic modelers wanting live demos of the OSI gauge; by day's end I'd booked follow-up calls with three North American utilities, two grid operators, and an inverter manufacturer that wants to embed the index. Most gratifying, though, were the off-record comments: a senior scientist told me, "This is the cleanest bridge I've seen between historian data and actionable BESS health," while a plant manager from Texas summed it up with a grin "You just gave me my weekends back.

Interviewer: Let's get into the tech. What tools and languages power this platform?

Drumil Joshi: Under the hood everything runs on a Python foundation: Pandas handles the heavy data-frame wrangling, scikit-learn plus SHAP drive the anomaly engine, and FastAPI exposes a lightweight REST layer. Raw telemetry flows from the OSIsoft PI Data Archive into an Airflow-orchestrated ETL that lands clean, millisecond-aligned tables in Delta Lake on Azure Databricks, where nightly notebooks retrain the Oscillation Severity Index models and push fresh weights back to production. The operator interface is a Dash application built with Plotly chosen for its real-time callbacks and pure-Python workflow which renders the OSI gauge, fleet heat-maps, FFT drill-downs, and the "Chat With SPC Helper" widget directly in the browser.

Interviewer: What does the future of this system look like?

Drumil Joshi: Next-up, we're turning the OSI engine from a fleet "stethoscope" into a fully digital-twin command centre. Over the next 12 months we'll layer in reinforcement-learning agents that can not only flag an emerging oscillation but also recommend and simulate the safest inverter or HVAC set-point tweaks before an operator clicks "apply." By mid-2026 the pipeline will ingest phasor-measurement-unit (PMU) streams and market prices alongside PI data, letting the model weigh grid-forming constraints and arbitrage opportunities in the same breath. We're also open sourcing the OSI specification so battery OEMs can bake it straight into firmware, while Southern Power pilots a SaaS version that any utility can spin up in under an hour on Azure or AWS. On the UI side, the Dash front end is getting a WebGL upgrade for million-point plots at 60 fps and a GPT-based co-pilot that drafts outage reports and NERC-CIP compliance notes on the fly. Ultimately, the goal is a self-optimizing loop: historian feeds → OSI scoring → prescriptive control suggestions → simulated impact → operator approval → automated dispatch-all audited, explainable, and export-ready for regulators and finance teams alike.

Interviewer: Last question. What advice would you give to others wanting to break into the AI + Energy domain?

Drumil Joshi: Start by mastering the fundamentals on both sides of the hyphen: build a rock-solid grasp of power-system physics everything from inverter control loops to capacity-market rules while sharpening the full Python-to-cloud ML stack, because flashy models collapse if you misread a single nameplate rating or timestamp. Next, chase "painkiller" use-cases, not "vitamin" demos: utilities will pay for lost-MWh recovery, uptime, and regulatory compliance long before they fund sci-fi prognostics. Third, treat data quality as a first-class engineering problem; spend as much time on historian tagging, unit conversions, and time-base alignment as on algorithm tuning, and automate those steps so your pipeline self-heals at 2 a.m. Finally, communicate like a bridge not a black box by pairing every model output with an explain-why narrative that control-room veteran's trust; the fastest way to adoption is when seasoned operators say, "That matches my gut, only sooner." Nail those four habits deep domain fluency, ruthless value focus, bulletproof data hygiene, and transparent storytelling and you'll find the AI + Energy door not just open but waiting for you to walk through.

Conclusion:

With a blend of AI, engineering, and visionary thinking, Drumil Joshi is setting a new benchmark in how we understand and manage clean energy infrastructure. His system isn't just a tool it's a transformation. As the world pivots toward sustainable energy, it's leaders like him who will ensure we do it smartly, securely, and at scale.