Siddhesh KambleVerified

I was one of the four members of Smoooth Operator, along with Tisha, Ravi, and Prathamesh. My main contribution was around alpha research and development, coming up with hypotheses, translating them into alpha expressions, and then iterating based on the backtest results.

As a team, we didn't have an extremely rigid division where one person only did one thing. We had different strengths, but the research process was collaborative. We would independently explore ideas, bring promising hypotheses to the team, test them, and then collectively decide which directions were worth pursuing further.

The biggest thing was that we treated the four of us as a research team rather than four individuals competing against each other.

Everyone could research and submit independently, so there was naturally some healthy competition. If someone found a strong alpha, the others wanted to understand why it worked and see whether they could find something equally good. But it was never about beating another teammate on the leaderboard.

We used LLMs extensively for research, understanding concepts, finding limitations in our hypotheses, and cross-verifying our ideas. They were also useful for exploring literature and challenging our assumptions.

Where AI let us down was that it could give very convincing but incorrect explanations or signals. So we used it as a research assistant, not as the final judge. Everything had to be validated through data and backtesting.

I don't think IQC has become purely a prompt-writing competition.

AI definitely lowers the barrier to exploring ideas. Someone who doesn't know every operator or piece of financial terminology can use an LLM to get up to speed much faster. In that sense, it can make research more accessible.

But the bottleneck moves rather than disappears. Once everyone has access to AI, simply asking an LLM to generate ten alphas isn't a meaningful advantage. The difficult part becomes deciding which hypotheses make economic sense, identifying what is actually novel, understanding why something works, and determining whether the result is robust rather than just a backtest artifact.

So I would say AI changes the skill set required for IQC, but it doesn't eliminate the need for quantitative thinking.

The biggest difference is the pace and depth of the research.

In a 48-hour hackathon, you have to optimize for speed. You identify the core problem quickly, build something that works, test it, and make sure you can present it convincingly before the deadline. You can't spend three weeks investigating whether an idea has a marginal improvement.

IQC is almost the opposite. Six months gives you enough time to throw away ideas, revisit them, build completely different research directions and gradually develop intuition.

The mindset I found most useful was treating IQC as a research marathon rather than one long hackathon. In a hackathon, you are trying to get to a working solution. In IQC, you are trying to build a repeatable research process that can keep producing good ideas for months.

I'd probably change the team format slightly. Since everyone can independently submit alphas, I'd make the team aspect more meaningful by introducing some mechanism that rewards collaboration and diversity of research within a team.

I think that would make the competition more interesting because you'd have to think not just about how good your own alphas are, but also about how well the four members complement each other.

I wouldn't immediately deploy one alpha with the full $1M.

The first thing I'd want to do is build a portfolio of genuinely complementary signals and allocate capital based on their risk characteristics. One of the biggest lessons from IQC for me was that a strong individual alpha isn't necessarily the same thing as a strong portfolio component.

What would worry me most is the gap between backtest and live performance. With real capital, transaction costs, market impact, regime changes and crowding become much more important. An alpha can look excellent historically and still behave very differently once real money is involved.

So before scaling it, I'd want to understand not just whether the alpha made money historically, but why it made money and under which market conditions that relationship breaks down.

I'd say Ravi. He handles pressure really well because he goes extremely deep into whatever topic he's working on. Even in stressful situations, he focuses on understanding the problem thoroughly rather than rushing to a solution. That depth gives him a lot of confidence when making decisions.

I think the advice that you should focus on making your expressions increasingly complicated is misleading.

Beginners often assume that if an alpha has more operators, more transformations and more layers, it must be more sophisticated. In reality, complexity can easily become a substitute for having a good hypothesis.

I'd rather start with a simple idea that has an intuitive reason for existing, test it properly, and then add complexity only if the data justifies it.

The other mistake is obsessing over one spectacular backtest. A single high-Sharpe result isn't necessarily valuable. Consistency, robustness, turnover and diversification matter much more when you're building a larger research portfolio.

I wouldn't say anyone in particular surprised me the most. All four of us brought different strengths to the team, and over the six months, we each contributed in different ways. I think that diversity in how we approached problems was one of the biggest strengths of the team.