DeepQDigital is an AI-native capital markets lab. We build AI trading systems and run them in our own digital asset trading business.
Build with usWe rebuilt the critical layers of the trading stack from the ground up, with AI as a first-class citizen at every level. Five systems, each built for one job.
The usual way
Most trading desks grow by adding teams. Each team works a narrow slice of market microstructure, builds its own signal, and the results are stitched into an ensemble.
It works, but it is slow, and cost rises with every strategy added.
01Trading infrastructure
The foundation. Pebble connects to centralized and on-chain venues, streams their market data in, and routes orders back out.
It is written in C++, runs in production on commodity cloud hardware, and has been benchmarked at 1 million messages per second. AI models are part of it from the start rather than attached later.
Making a trading system fast is the easy part. Venues send fills before acknowledgements, resend old messages, skip sequence numbers and go quiet without disconnecting. None of that crashes a system; it shows up later as a position that no longer matches the venue. We know where it breaks, which is why AI tools make us faster rather than riskier.
02Alpha generation
An AI model of the market, built on transformer and world-model techniques.
A transformer reads live order books across assets and venues together and compresses them into one picture of market state. A world model then rolls that state forward to predict whole order books several steps ahead, not just the next price.
Years on trading floors shape what it learns from and which noise it ignores.
03Reinforcement learning
Hive decides what to do. It frames each strategy as a reinforcement learning problem: the state of the market, the actions available, and the reward that follows.
Any single trade is noisy. Hive learns behaviors that win across many of them, starting from proven, expert-driven strategies and searching the space of decisions in a way grid search and manual tuning struggle to match.
04Agentic workflow automation
Metis does the research work. It runs backtests, validates data, analyzes fills and tests new ideas overnight, while the team is offline.
It follows our research standards and knows our systems, data and protocols. Nothing it produces reaches production without human review.
05Accounting and reporting
Our fund accounting system. Each day it collects balances and trades from every venue, wallet and bank, checks them against the books, and records each trade in a double-entry journal.
A person approves the close. Ledger then allocates results to each investor’s capital account and produces NAV, financial statements and investor reporting from the same records. The back office runs as a system, not a team.
The live loop. Runs continuously. Pebble brings in data, Prometheus predicts, Hive acts, Pebble executes, and Ledger books the result.
The research loop. Runs around it. Metis studies the results and proposes improvements, which go live after human review.
Read: An AI-native trading stack, built from first principlesThe systems we run every day are available to firms building their own platform. Adopt one layer or the whole stack, and build it with people who have operated electronic trading desks.
How we build with you2016
We met while organizing a teach-in at Deutsche Bank on artificial intelligence and cryptocurrency. The consensus view at the time was that neither technology would meaningfully improve upon existing infrastructure.
2022
We left Deutsche Bank and Bank of America to build DeepQDigital from the ground up. Digital asset markets became the laboratory to reimagine the full trading stack with AI-native systems, while keeping the best practices of decades serving institutional clients on Wall Street.
Today
The infrastructure is live. We actively trade spot, perpetual futures and prediction markets. As blockchain infrastructure blurs the line between digital and traditional asset classes, we believe this AI-native framework extends to every asset class.
The name is an homage to the Deep Q reinforcement learning algorithm.
Chief Executive Officer
FX electronic trading at Deutsche Bank and Bank of America. MBA, Columbia Business School.
LinkedIn profile of Carlos Zendejas (opens in a new tab)
Chief Investment Officer
U.S. Treasuries electronic trading at Deutsche Bank. PhD, University of Warwick, with doctoral research at CERN on the Large Hadron Collider.
LinkedIn profile of Matt Reid (opens in a new tab)
Chief Science Officer
PhD in Theoretical and Mathematical Physics, University of Hamburg. MASt in Applied Mathematics, University of Cambridge.
LinkedIn profile of Niklas Henke (opens in a new tab)Live P&L in adversarial markets tells our story more credibly than internal benchmarks or research evals.
We operate a systematic, market-neutral trading business focused on microstructural opportunities in digital assets. Our edge is execution, not directional bets. We trade spot, perpetual futures and prediction markets across major centralized and on-chain venues.
DeepQDigital is an AI-native capital markets lab based in Greenwich, Connecticut. It was founded in 2022 by former Deutsche Bank and Bank of America electronic traders. It builds AI trading systems and runs them in its own systematic, market-neutral digital asset trading business.
DeepQ trades spot, perpetual futures and prediction markets across major centralized and on-chain digital asset venues. The firm believes the same AI-native framework can be extended to all asset classes as blockchain infrastructure blurs the line between digital and traditional assets.
The strategy is offered to qualified institutional investors through a private fund vehicle. Qualified institutions can write to investors@deepqdigital.com. Nothing on this site is an offer to sell or a solicitation of an offer to buy securities.
Yes. DeepQ works with firms that want to build their own trading platform, from a single layer such as exchange connectivity or fund accounting to a full AI-native stack. Engagements use components DeepQ runs in production. Write to partnerships@deepqdigital.com.
We respond to thoughtful inbound messages. We do not respond to unsolicited sales pitches or recruiting agencies.
