The Minds Behind the Models: Kushpreet Singh

Kushpreet Singh: The Statistician Trading America's Least Documented Market

Kushpreet Singh has a background that reads like a purist's résumé: a bachelor's in mathematics and computer science from Chennai Mathematical Institute, a master's in mathematics and statistics at the Indian Statistical Institute in New Delhi, and an M.Stat in mathematical statistics and probability at ISI Kolkata. Modelling, he says, always interested him — but it was a series of freelance data science and quant projects after his master's that pulled him properly into the field.

"Modeling always interested me, and after my master's I picked up some freelance data science/quant projects — that's what got me hooked. I also started competing in online quant competitions, which really shaped my sense of what methods actually work versus what just looks good on paper."

That last line is the thread running through everything Kushpreet does. He's not interested in what looks good on paper. He's interested in what survives contact with a live market.

Today he works full‑time as a quant at Progrid Analytics, a small US‑based energy trading firm. His team runs live algorithmic strategies across multiple US power markets, and he's a core member of the research group — responsible for strategy R&D, production deployment, and expansion into new grids. He's independently designed and deployed a probabilistic ML‑based strategy end‑to‑end: feature engineering, modelling, portfolio construction, and production code, live since November 2025.

He also co‑developed a second live strategy on a different product within the same markets, and handles live monitoring and performance diagnostics. For a small team in a market with almost no published research, that means a lot of the design work has to happen from first principles.

Energy Markets: Thrilling, Volatile, and Barely Documented

Kushpreet's day job sits at the frontier of what's written down. There's very little published research on virtual trading in energy markets, which means strategy design, risk management, and monitoring tooling all had to be built from scratch.

"I work full‑time as a quant at a small US‑based energy trading firm — one of the newest and most volatile markets out there right now. There's very little published research on virtual trading in energy markets, so a lot of the strategy design, risk management, and monitoring tools had to be built from scratch. I learned by actually trading, then iterating based on live PnL rather than backtests alone. It's not for the faint of heart — weather‑driven volatility keeps things thrilling."

That phrase — iterating based on live PnL rather than backtests alone — captures the professional discipline that separates a practitioner from a competitor. It's the same instinct we've seen across this series: Mathurin and Alok both built evaluators calibrated against reality rather than trusting their local numbers. Kushpreet lives in that reality every day.

Why Point Forecasts Don't Cut It

When asked what tool or technique he reaches for most, Kushpreet pointed to something more foundational than a library: distributional forecasting.

"Early on I realized point forecasts don't cut it in energy markets — prices have fat tails, so you need distributional forecasts instead. Over the past year, probabilistic forecasting via ML has picked up a lot of attention, which only confirmed that instinct."

Fat tails are a recurring theme across quantitative finance, and energy markets are one of the places where they bite hardest. A model that predicts the expected price is useless when the market is defined by rare, extreme moves driven by weather, outages, and grid constraints.

But his bigger lesson was structural, not statistical:

"The bigger lesson though: trading isn't just fitting a regression model for returns. You need a downstream layer that takes those model outputs and decides how much to trade and when — that's where the real edge is."

This is a subtle and important point. A signal is not a strategy. Between the prediction and the P&L sits a decision layer — sizing, timing, risk limits, regime awareness — and that layer often matters more than the model itself. It's the same insight that shows up repeatedly across these profiles: the infrastructure around the model is the edge.

Regime‑Conditional Thinking and a Hierarchical Architecture

For AlphaNova, Kushpreet's approach leaned on distributional regression, novel feature engineering, and regime‑conditional models — a direct import from his energy trading experience, where signals behave very differently depending on market state.

He also described an idea he's currently exploring:

"One idea I'm exploring is a hierarchical setup: a slow‑moving base alpha combined with a fast controller layer that reacts to recent market conditions."

The architecture is appealing precisely because it mirrors what trading actually looks like: a durable, slower‑moving view of the market, modulated in real time by a layer that responds to short‑term conditions. He's honest that he hasn't fully implemented or tested it yet:

"Haven't fully implemented or tested these yet, but I plan to try them out in the current competition and keep pushing on these ideas in future ones too."

That candour is worth noting. Most competition profiles are written after the fact, with the benefit of knowing what worked. Kushpreet is describing his process in progress — the way a real researcher talks about ideas that are still being shaped.

Agentic AI: "Implementation Is Almost Free"

On how he uses AI, Kushpreet was unambiguous: agentic tools have completely transformed how he works.

"Agentic AI has completely transformed my workflow. As a math person, it's a huge boon — I can now think and solve problems at a higher level of abstraction while implementation becomes almost free and fast. I've genuinely entered a kind of flow state with these tools and can work on multiple projects in parallel now."

And this is precisely why he's taking competition platforms more seriously:

"That's exactly why I'm taking platforms like AlphaNova much more seriously — there's no longer an implementation bottleneck between having an idea and testing it live."

This is the same shift we've documented across the series. Taro never learned to code formally and still cracked the top ten. Alok tested nine signal variants in a single evening. Saksham generated 28 candidate signals in two weeks. When implementation stops being the bottleneck, the differentiator becomes judgment — what to build, what to believe, and what to throw away.

On AlphaNova and Open Research

Kushpreet had some generous words for our co‑founder, Marc Nunes, and about what AlphaNova is trying to build.

"Marc is brilliant, funny, and remarkably generous — he's putting out all these gems for free to the quant community. I've had a few interactions with him over Discord and WhatsApp calls, and quant is a notoriously secretive business, so it's rare to see someone as illustrious as him share this much knowledge openly. I always want to support ambitious ideas with huge potential like AlphaNova — it feels like a genuine win‑win for the platform and the competitors alike."

If you're interested in the kind of research he's referring to, our article on signal diversity, correlation, and the geometry of forecast libraries covers three papers recently published by Marc on the structure and limits of signal collections.

What We Can All Learn

Kushpreet's profile offers three lessons that extend well beyond energy markets:

Point forecasts are rarely enough. In fat‑tailed markets, distributional forecasts carry information that a single expected value throws away.
The downstream layer is where the edge lives. A model output isn't a strategy. Sizing, timing, and risk decisions often dominate the outcome.
Agentic AI removes the implementation bottleneck. When building and testing is nearly free, the scarce resource becomes the quality of your ideas and your judgment about which ones to trust.

That combination — statistical rigour, live‑market discipline, and honest reporting of ideas still in progress — is exactly what the AlphaNova community is built on.

Stay tuned for more profiles from the AlphaNova community. If you'd like to be featured, reach out—we'd love to share your story.