Quantitative systems builder · research · validation · execution · risk
UCL MEng · Year in Industry @ SLB · London
I build systems for making decisions under uncertainty, drawing on a background in mechanical engineering, control systems, embedded sensing and quantitative finance.
I graduated from University College London in 2026 with an MEng in Mechanical Engineering with Year in Industry, achieving Upper Second-Class Honours (2:1). I am particularly interested in time-series research, state-space models, systematic trading, execution and risk.
Open to graduate and early-career opportunities in quantitative development, trading systems, research engineering and adjacent market-technology roles.
QuantSilico is an AI-assisted quantitative research, validation and deployment framework. It connects data preparation, signal research, backtesting, configuration promotion, risk controls, MT5 execution, telemetry and post-run review.
research -> validate -> deploy -> monitor -> review
The aim is not simply to generate more strategies. The system is designed to reject weak ideas, retain evidence from failed experiments and keep deterministic risk gates in authority over deployment.
AI assists research and critique; deterministic code and hard risk controls govern execution.
The strongest result is not that every idea worked; it is that the system could identify which ideas should not be deployed.
The product direction extends beyond the original competition towards broader real-market research validation and governed deployment. The commercial platform is still under active development.
During the Syphonix Model to Market competition, I operated a config-driven systematic strategy stack on a simulated $1 million account across 15 FX, metals and crypto instruments. The solo entry finished 19th out of 440 participants, inside the top-25 prize bracket.
→ quantsilico-model-to-market-competition · portfolio case study
Sole-author research submission on geopolitical transmission into oil markets via insurance, shipping and product constraints.
→ quantsilico-onyx-future-of-energy-trading-competition · Submitted; no public ranking was issued.
Twelve binary forecasts with measurable thresholds, resolution dates, authoritative resolvers and a causal framework spanning AI investment, energy systems, strategic inputs and industrial policy.
→ quantsilico-forecasting-the-future-2026-competition
Applying real-time state estimation to pairs trading: dynamic hedge-ratio estimation, spread construction, stationarity analysis, cost-aware entry and exit logic, and walk-forward comparison with static-hedge approaches.
Custom Gym-style environment work covering Monte Carlo methods, temporal-difference learning, Q-learning and SARSA, with uncertainty, rewards and transaction-like costs.
As software and electronics lead in a five-person multidisciplinary team, I developed the embedded sensing chain, edge-processing services, local raw-data storage, 4G/LTE store-and-forward telemetry and dashboard path for a coastal-monitoring prototype.
The supervised two-hour Brighton field window recorded locally while maintaining live position and system reporting. The full-window record reported 99.9% upload delivery, with 52 files acknowledged and none pending. Power measurements supported an estimated continuous endurance of approximately 23 hours at the tested load; that was a projection, not the field-test duration.
→ coastal-buoy-monitoring-platform · portfolio case study
Individual computational study connecting protein representation, target-conditioned molecular generation, model-based interaction scoring and cheminformatics property screening for an FGFR2 case study in intrahepatic cholangiocarcinoma.
The available analysed archive contains 100 valid molecular records; 17 met the project-defined drug_score_total >= 0.7 screen. These are generated structures and model-based prioritisation signals, not experimentally validated drug candidates.
→ fgfr2-generative-drug-discovery · portfolio case study
At SLB, I worked on real-time orientation and control problems for measurement-while-drilling systems. This included Extended and Square Root Cubature Kalman filtering, Python tooling for diagnosing data-conversion defects, and validation of a motor/control digital twin against field traces.
- Approximately 15% reduction in magnetic-bias error on replayed well data
- Approximately 45% reduction in debugging time through Python automation
- Digital-twin validation within approximately 2% RMS of field traces
This experience is the engineering foundation for my work on state estimation, noisy data and quantitative systems.
A promising backtest is the beginning of the review process, not the end.
Models can accelerate analysis and critique, while deterministic risk controls retain authority.
Knowing why an idea should not be deployed is part of building a robust research system.
Languages and engineering
Quantitative research
Markets and infrastructure
Competitive powerlifter. I like problems that do not care how clever you are—only whether the bar moves.
Outside engineering, I powerlift. The unofficial stack is Python, C++, progressive overload and a completely ordinary amount of creatine.
