The Minds Behind the Models: Akshita Srivastava
Akshita Srivastava: The Actuarial Mindset and the "Volume Over Polish" Philosophy
Akshita Srivastava is an Actuarial Science and Mathematics undergraduate at the University of Manchester, a WorldQuant BRAIN Research Consultant, and a two-time hackathon winner (including 1st place at QuantiHack ’26 and the AMD Developer Hackathon). She’s an EDIA Prize recipient, a Silver Medalist at the South-East Asian Mathematical Olympiad (SEAMO), an IOQM Merit Holder, and—somehow—she has also discovered an asteroid as part of an astronomy research program.
But what makes her profile truly stand out isn't just the list of accolades. It’s her relentless focus on finding the truth, even when the truth is inconvenient.
Her entry into quantitative finance came through her peers.
"A lot of my friends were into quant, so I kept hearing about it. Then I found out about QuantiHack, competed in the initial rounds and qualified for the finals, and we won 1st place at the London finals with my team, Team APEX. That’s when it clicked for me. After that I went deeper: IMC Prosperity, where I finished in the top 0.46% globally and 10th in the UK, and WorldQuant BRAIN, where I’m now a Research Consultant."
That rapid trajectory—from a student hackathon to the top 0.5% of global competitors—is impressive. But it is her approach to signal design that offers the most valuable lessons.
"Volume Over Polish"
In both the AlphaNova competition and her work at WorldQuant, Akshita prioritizes quantity of independent ideas over hyper-optimizing a single strategy.
"I build signals the same way on every platform. I start with a simple economic idea, often fundamentals-based, and change one thing at a time while logging every result, so I know what actually helped. WorldQuant’s IQC taught me to value volume over polish. Early on I over-tuned single signals, and I now prefer several decent, uncorrelated signals across different data types."
Her primary tool for this is ts_rank applied to fundamental ratios like earnings yield. Instead of looking at the raw value (which can be distorted by broad market shifts), ts_rank ranks each stock against its own history.
She also relies heavily on parameter sweeping, testing a signal across different decay, universe, and neutralization settings rather than tuning a single version forever. But she adds a crucial validation step:
"Then I look at how it did in 2024 and 2025 on its own, because a signal that only worked in older years isn’t worth keeping."
The FDA Recall Test: Trying to Break Her Own Signals
The most striking part of Akshita’s methodology is her commitment to falsification. She doesn't just test her signals; she actively tries to break them.
During QuantiHack, she and her team tested two of their signals against real FDA recalls. The results were close to chance.
"At QuantiHack we tested two of our signals against real FDA recalls and got results close to chance, so we removed the accuracy claims and published the real numbers."
That is intellectual honesty in its purest form. It is incredibly easy to tweak a model until it fits the historical data, and even easier to publish the wins while sweeping the losses under the rug. Akshita did the opposite. She removed the claims and published the truth.
"I also try to break my own signals before trusting them."
This instinct is the same one that drove Ismam to retire an underperforming architecture entirely, and Vivaan to write a pre-submission tester that scrambles the future and shuffles data to catch overfitting.
Research Without a Known Answer
Currently, Akshita is working on part-time research in quantitative finance and machine learning with academics in Europe. For her, the shift from competitions to research represents a fundamental change in how she works.
"I’m excited about it because it’s on topics I’m still learning the maths for, so it pushes me to understand things properly instead of just running code. It’s also the first time I’ve worked on something with no known answer to check against, which is a different feeling from a competition."
In a competition, the leaderboard tells you if you are right. In research, you have to build the definition of "right" yourself. That is where the actuarial mindset—managing risk under extreme uncertainty—becomes a superpower.
AI as a Devil’s Advocate
Akshita uses Claude heavily, but her use case is deeply focused on intellectual rigor.
"I use it to build and debug code for hackathon projects and competition bots, to learn maths step by step, and as a devil’s advocate I give it an idea and ask it to find the holes before I spend hours on it. I don’t take its output as given. I check numbers against the actual results, and I’ve caught real errors that way."
She uses AI not to make the work easier, but to make her own thinking harder to attack.
"For me it works best when it makes my own thinking harder to attack, not easier to skip."
That is a masterclass in how to use LLMs for quantitative research. The AI isn't there to give you the answer; it's there to pressure-test your assumptions before you commit time and capital to them.
What We Can All Learn
Akshita’s profile offers three distinct lessons for building robust quantitative strategies:
- Value volume over polish. Instead of over-tuning one signal until it looks perfect on a backtest, build several decent, uncorrelated signals across different data types.
- Test against real-world events. Don't just validate against a generic target variable. Test your signal against real-world outcomes (like FDA recalls or earnings surprises). If it fails, report the truth and move on.
- Use AI as a devil's advocate. Don't ask the AI to find alpha. Ask it to find the holes in your hypothesis before you spend hours coding it.
That combination of actuarial discipline, intellectual honesty, and relentless curiosity 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.