Real-time market making demands infrastructure that can absorb vast amounts of data, process signals in milliseconds, and still leave room for human oversight. In this episode of Data in the AI Era, Chris Dale and Nikos Tsoskounoglou, co-heads of quantitative trading at ADSS, explain how they’ve built and scaled a system that manages exactly that challenge.
Episode overview
The discussion begins with their journey: starting as a two-person team with a minimal setup — Python and limited datasets — and evolving to a kdb+-based environment ingesting over a billion ticks per day. Their stack runs analytics in memory, delivers real-time signals to both algos and traders, and enforces strict controls to prevent models from overstepping. The approach balances automation with human supervision: algorithms handle sub-millisecond decisions, while humans intervene in edge cases, similar to a pilot stepping in when autopilot goes out of bounds.
“Nikos and I like to think of the quant platform we’ve built on top of kdb+ as the brain of our trading infrastructure. It makes real-time decisions on pricing, on when to go to market, and on how to manage risk. All of that logic resides within the platform we are responsible for.” – Chris Dale, ADSS
Chris and Nikos go on to highlight the metrics that matter. Profitability and spread retention remain core KPIs, but strategy evaluation depends on multi-metric backtesting across thousands of candidate models. They stress that precision isn’t just about tighter spreads; it’s about designing hedging strategies that open new revenue channels while maintaining fair client pricing.
A defining part of their progress has been the way they structured the team. Researchers, engineers, and traders collaborate closely, with ideas flowing quickly between model design, implementation, and execution. Crucially, democratization of tools has broadened participation. With PyKX, Python users across the desk can query and analyze data directly in kdb+, reducing bottlenecks and enabling more voices to contribute to research and decision-making. This accessibility has created a more technical, data-driven trading desk where innovation comes from all sides.
Finally, they look forward to expanding their use of traditional machine learning, scaling feature-rich models, and cautiously exploring agentic AI frameworks. For them, the future is about innovating without hype: leveraging trusted data foundations and proven ML techniques to drive incremental improvements in execution quality and risk management.
Key takeaways
1. Precision is more than speed
Latency is critical in market making, but precision extends beyond execution time. It means offering competitive prices that reflect risk, inventory, and hedging capacity while maintaining fairness for clients. Precision also involves building new revenue channels, such as designing hedging algorithms that allow spreads to tighten without sacrificing profitability. Speed matters, but the real challenge is delivering the right price at the right time while managing exposure and sustaining returns.
2. Infrastructure dictates possibility
The ADSS team progressed from a basic Python setup to a kdb+ environment ingesting over a billion ticks per day. This shift enabled real-time analytics, memory-resident models, and latency-sensitive pipelines tuned for specific use cases. Infrastructure defines the frontier of what can be achieved. Without scalable ingestion, high-speed retrieval, and low-latency architecture, even well-designed models cannot perform in production. A carefully engineered stack is what turns theory into live trading capability.
3. Humans and machines in balance
Automation drives the bulk of ADSS’s market-making stack, but strict limits prevent algorithms from exceeding defined parameters. Hedging models operate only within set bounds, and unusual price data prompts human review. Chris compares this to autopilot in aviation: machines handle sub-millisecond events, while humans step in when operations move outside expected ranges. Resilience comes from bounded autonomy. Algorithms provide scale and speed, but human oversight ensures fairness, prevents breakdowns, and preserves trust with clients and regulators.
4. Culture of curiosity and democratization
ADSS built a culture where ideas move quickly from concept to production, supported by open collaboration and tolerance for experimentation. Researchers, engineers, and traders exchange feedback daily, creating short loops between theory, implementation, and live trading outcomes. An important enabler has been the democratization of tools. With PyKX, Python users across the desk can query and analyze kdb+ data directly, reducing reliance on specialists and giving more people the ability to contribute insights. This accessibility means traders, researchers, and engineers are all closer to the same data, accelerating innovation and ensuring strategies evolve as fast as the market itself.
5. Strong data foundations matter most
High-quality data underpins everything the team does. Early investment in capturing and validating billions of daily ticks gave them confidence in analytics, backtests, and live decisions. Reliable data enables scaling of models, evaluation of thousands of strategies, and safe exploration of machine learning or unstructured sources. Sophisticated models are only as good as their inputs. Strong governance of data compounds over time, unlocking innovation while controlling risk
Key quotes
“We engaged with KX to build a tech stack that gave us far more quantitative power — in terms of the research we could do, the data we could capture, and the problems we could investigate. That was a game changer for us.” – Chris Dale, ADSS
“PyKX has taken democratization to the next level for us. q is a tough language to learn, especially if you don’t have a technical background, but Python is a far more common skillset. PyKX makes it much easier to couple a Python environment with data in kdb+, and that has been tremendously powerful for us.” – Chris Dale, ADSS
“Today’s trading desk is far more technical than it was in the nineties. Dealers can write kdb+ code (q) , work in Python, and analyze data for themselves.” – Nikos Tsoskounoglou, ADSS
Want to learn more? Explore how the high-performance analytical database for the AI era can help you ask better questions and act faster on your data.
Be first in line to hear our next podcast by subscribing to our YouTube channel, Apple Podcasts, Spotify, or your chosen podcast platform. If you have any questions you would like us to tackle in a future episode or want to provide some feedback, reach out to the team at podcast@kx.com.
