On October 21, 2025, Quantopian hosted an inspiring webinar featuring Jiri Pik, Ernest (“Ernie”) P. Chan, and Jared Broad – three co-authors of the new Wiley book Hands-On AI Trading with Python, QuantConnect, and AWS. Moderated by Prof. Esfan Haghverdi, this conversation delved into practical applications of AI in finance, showcased a live trading strategy demo on QuantConnect, and shared hard-won wisdom on leveraging machine learning in quantitative trading.
Below, we recap the key insights and takeaways from this event, which mixed professional depth with motivating encouragement for aspiring quant traders.
Bridging AI and Quant Finance – The Vision Behind Hands-On AI Trading
Jiri Pik opened by explaining the genesis of the book. The authors saw a gap in the market: while many texts discuss AI in trading, their coverage of methods is often scattered. “Our goal was to collect all relevant AI methods into one comprehensive piece and focus entirely on teaching intuition and insights into how each AI method works,” Jiri noted. Instead of writing yet another theory-heavy volume, they set out to bridge the gap between artificial intelligence and quantitative trading by providing a single, practical guide spanning classical models to deep learning, LLMs, and reinforcement learning.
Crucially, the team chose QuantConnect as the backbone for all examples. “Most other books spend half their content on setting up infrastructure and obtaining correct market data. By using QuantConnect, we eliminated that problem and could focus on strategy development instead,” Jiri explained. QuantConnect’s cloud-based platform offers ready access to clean market data and a robust backtesting engine, allowing the authors to dive straight into strategy implementation. They identified key AI techniques used in modern finance, built representative examples (20+ in total), and organized them by increasing complexity so that readers can start from scratch and progressively master each technique. This structure – from foundations to advanced applications – reflects the book’s mission to turn theory into hands-on skills.
Ernest Chan recounted that when invited to contribute, he was excited to consolidate his insights in a cohesive way. At his company PredictNow.ai, new AI trading techniques were being developed across blogs and papers; having them in one working code repository was ideal. Chan also injected a note of realism about AI’s current abilities in trading. In his experience, “tools like ChatGPT for designing trading strategies produce flaky results—they’re not robust enough across different market regimes”. AI excels at supporting tasks – generating reports, explaining code, optimizing parameters – but “for designing a complete trading strategy from scratch, I wouldn’t rely solely on AI”. This theme of AI as powerful assistant, not autonomous guru, would recur throughout the discussion.
Jared Broad, founder of QuantConnect, added that this was his first foray into authoring a book, and the project took about 1.5 years from conception to publication. It was a team effort aligning diverse expertise (quant finance, AI, cloud infrastructure) into a unified vision. The result is a book that doesn’t just teach theory – it provides working code and a platform to try everything yourself, embodying the “hands-on” spirit.
QuantConnect Demo: Piotroski F-Score Strategy in Action
To illustrate practical AI trading in real time, Jared Broad led a live demo on the QuantConnect platform. The strategy chosen was the Piotroski F-Score strategy – a classic fundamental stock-picking method from about 25 years ago that scores companies based on financial strength.
Strategy Concept: The Piotroski F-Score assigns points to stocks based on multiple financial criteria (profitability, leverage, operating efficiency, etc.). In this demo strategy:
- Only stocks with an F-Score above 7 (out of 9) are included in the portfolio.
- The portfolio is rebalanced monthly (30-day holding periods).
- Positions are weighted by sector and liquidity filters are applied (e.g. minimum price and volume).
Using QuantConnect’s Python API, Jared showed how quickly one can implement this logic. The platform provides a browser-based IDE and event-driven backtester – you write the algorithm (e.g., define a universe of stocks, request fundamental data, rank by F-Score) and QuantConnect handles the rest (data retrieval, corporate actions, performance tracking). The backtest ran across the full historical dataset, producing an equity curve, drawdown stats, and trade logs.
Parameter Optimization: A highlight of the demo was QuantConnect’s integrated parameter optimizer. “Most machine learning techniques have hyperparameters. The beauty of QuantConnect is that it includes an optimizer,” Dr. Chan had noted. Jared demonstrated this by tuning two key strategy parameters:
- F-Score Threshold – the minimum score for inclusion (tested values 7 vs. 8).
- Universe Size – the number of top-scoring stocks to hold.
The platform automatically ran backtests for each parameter combination and generated a heatmap of results. The outcome was illuminating: a threshold of 7 produced a Sharpe ratio around 0.8–0.9, whereas tightening the threshold to 8 caused Sharpe to drop to ~0.4. This stark contrast suggested that the stricter criterion was likely overfitting – a higher score threshold didn’t generalize well, a subtle insight that might have been missed without systematic optimization. “QuantConnect’s integrated optimizer lets you test parameter sensitivity instantly without rebuilding infrastructure – a capability that would take years to develop independently,” Jared emphasized. In other words, what used to require custom code and lengthy experimentation can now be done in a few clicks, allowing quants to focus on insights over infrastructure.
This demo underscored how AI and cloud technology democratize algorithmic trading. A strategy from academic literature was prototyped, tested, and improved within minutes. The message to readers and viewers was clear: these tools empower you to iterate quickly, discover what works, and learn from what doesn’t in a risk-free environment. It’s a practical sandbox for turning ideas into tested strategies – exactly the hands-on experience the book advocates.
Harnessing AI Tools Wisely: From Generative AI to Reinforcement Learning
A significant portion of the conversation explored how AI techniques are reshaping trading, and the note was one of cautious optimism. The speakers acknowledged the buzz around Generative AI and large language models (LLMs) in finance. Jiri Pik mentioned that the book includes a chapter on generative AI for trading, covering architectures and prompt engineering basics. However, both he and Ernie Chan cautioned that LLMs are not (yet) one-stop strategy designers. Jiri shared that in his experiments using ChatGPT and Claude to design strategies, “results are usually flaky – not robust across different market regimes and nuances”, reinforcing Chan’s earlier point. AI excels at augmenting human quants – e.g. writing boilerplate code, summarizing research, or suggesting improvements – but human judgment remains critical. As Dr. Chan put it, LLMs can show “flashes of brilliance and utter absurdity” in the trading domain, and *if you’re not already an expert, you “can’t distinguish genius from nonsense” in those outputs. In short, quants should approach these shiny new AI assistants as powerful “copilots” rather than autonomous pilots.
Jared Broad illustrated how QuantConnect responded to the LLM trend. As more users tried ChatGPT for coding strategies, they found raw prompts often failed (only ~25% would run successfully). So his team built “Workflows” – structured micro-prompt sequences with error-checking loops – to guide the AI step by step. By programmatically feeding back compiler errors and backtest results for iterative improvement, they boosted the success rate of AI-generated algorithms to ~75%. He also mentioned new Model Context Protocols (MCPs) – essentially autonomous agent frameworks that can devise and execute their own plan for a task with minimal human intervention. These advances hint at a future where AI might handle more of the strategy development, but for now the consensus was that human expertise and oversight are indispensable.
The panel also ventured into reinforcement learning (RL), another cutting-edge AI topic covered in the book (e.g. “Better Hedging with Reinforcement Learning” in Chapter 7). Dr. Chan, a physicist-turned-quant, offered a nuanced take on RL’s place in finance. RL is superb for problems where no clear “right answer” exists – situations where you learn by trial-and-error reward maximization rather than predicting a known outcome. Trading can be such a domain, and interestingly, finance avoids one of RL’s hardest hurdles: the lack of counterfactuals. In many real-world tasks, you can’t easily ask “what if we had taken a different action?” since history can’t be replayed. But markets are more like a simulator – you can ask “What if I made a different trade?” and backtest it. This means finance is fertile ground for RL, as you can generate the alternative scenarios needed for learning.
That said, Chan warned that RL in practice is very resource-intensive and finicky. Unlike supervised learning, which is straightforward to train and evaluate, RL’s outcomes have high variance – two training runs can yield wildly different behaviors. Achieving reliable performance often requires running hundreds of parallel agents and averaging their policies, demanding enormous computing power. Even after spending six months on a deep RL project, Chan found the results were “not substantially better than simpler methods”. His message wasn’t to dismiss RL outright, but to set realistic expectations: RL can unlock new strategies (especially where traditional approaches fall short), but you must be prepared to invest time, compute, and careful tuning, with no guaranteed jackpot at the end. It’s a cutting-edge tool best approached with both enthusiasm and healthy skepticism.
Fostering Curiosity and Continuous Learning in Quant Finance
Despite covering advanced topics, the tone of the webinar was overwhelmingly encouraging and empowering. The underlying theme was that any motivated individual can start exploring AI in trading today, thanks to accessible resources like the book and platforms like QuantConnect. The guests repeatedly stressed the importance of hands-on experimentation and community learning over passive consumption.
Jared Broad highlighted that QuantConnect now hosts hundreds of example algorithms and a vibrant user community. “For those interested in these topics, QuantConnect has hundreds of examples and a vibrant community. The best way to learn is by doing,” he said. In other words, reading and theory must be complemented by rolling up your sleeves – clone an example, tweak a parameter, run a backtest, and see what happens. Each iteration deepens understanding, and even failures teach invaluable lessons in the nuances of markets and model behavior.
Jiri Pik echoed this sentiment in his closing remarks. He encouraged the audience to approach AI in trading with a spirit of play and inquiry: “We encourage curiosity. Modify examples, combine techniques, and explore ‘what-if’ scenarios,” Jiri advised. The infrastructure for experimentation – cloud data, open-source libraries, online forums – is more powerful than ever. By asking “what if?” and then actually testing it, aspiring quants can discover insights that no textbook or degree alone will provide. Jiri’s enthusiasm underlined that innovation in finance now often comes from tinkerers and cross-disciplinary thinkers who aren’t afraid to break things (in simulation) and learn from the process.
Finally, the panel expressed gratitude to the community and reinforced the value of collaboration. The book itself was a collaboration between five authors with backgrounds ranging from physics and engineering to hedge funds and cloud platforms – an example of how diverse perspectives can create something greater than the sum of its parts. “The book wouldn’t exist without [the] team’s diverse expertise… Thank you to the community for making this possible,” said Chan, reflecting on the supportive ecosystem around quantitative finance. In the same cooperative spirit, QuantConnect and Quantopian have fostered communities where newcomers can interact with experts, share ideas, and accelerate their learning.
In summary, the Quantopian Book Conversation with Jiri Pik, Ernie Chan, and Jared Broad offered a treasure trove of insights: a clear vision for how AI can be taught and applied in finance, a live demonstration of that vision via a fundamental AI-driven strategy, and balanced perspectives on emerging tools like LLMs and RL. Most importantly, it delivered a message of inspiration – that with curiosity, diligence, and the right tools, anyone passionate about quantitative trading can participate in this rapidly evolving field.
Hands-On AI Trading with Python, QuantConnect, and AWS (Wiley, 2025) is more than just a book title – it’s an invitation to actually get hands-on and start building your own algorithmsjiripik.com. Whether you’re a student, a developer, or a finance professional, now is the time to explore these techniques, leverage community platforms, and perhaps turn some of those “what if we tried this?” ideas into reality.
Special thanks to Quantopian and Prof. Esfan Haghverdi for hosting this insightful session and to the speakers for sharing their experience. The webinar exemplified the best of our community – professionals and enthusiasts coming together to push the frontier of AI in trading, while openly sharing knowledge to educate and motivate others.
Ready to dive deeper? Check out the book Hands-On AI Trading with Python, QuantConnect, and AWS and visit the QuantConnect platform to start experimenting with AI trading strategies yourself. The future of finance is being coded today – and these tools and insights can help you become a part of it.
