Role Author/Stack Zybo Z7-20, Verilog, Vivado 2019.1, C++ (Xilinx SDK)/Date Apr 2023/Track Systems
Problem. Detect edges in live camera video in real time, on hardware rather than in software.
Method. Edge filters with Prewitt and Sobel weights, hand-written in Verilog and inserted into the HDMI output stream of Digilent’s Zybo Z7-20 Pcam 5C demonstration design.
Result. Timing is not met at the 148.5 MHz pixel clock: worst negative slack −92.2 ns (Prewitt) and −90.3 ns (Sobel) in the committed Vivado 2019.1 routed reports
Role Author/Stack C++, GoogleTest/Date Mar 2024/Track Quant, Systems
Problem. Accept and match several order types in one book, including good-till-cancelled and market orders, with behaviour that can be checked rather than trusted.
Method. Order types implemented in C++ (Good-Till-Cancel, Fill-and-Kill, Fill-or-Kill, Good-For-Day, Market) on a std::map book; GoogleTest cases that pin price priority and crossing; a seeded benchmark of latency, throughput and allocations over 10 repetitions.
Result. p50 / p99 per call: add 208 / 625 ns, cancel 125 / 333 ns, match 1,667–1,708 / 4,062–4,125 ns (OrderbookSim on an Apple M1 Pro, median of 10 repetitions; includes history and neural-network bookkeeping)
Problem. Price European options correctly, and show the error alongside the answer.
Method. Black–Scholes prices and Greeks checked against finite differences; an implied-volatility solver (Newton with a bisection fallback); a seeded Monte Carlo pricer with a 95% interval; put–call parity as a unit test. All in C++17.
Result. Monte Carlo RMS error falls from 1.446 at 100 paths to 0.0044 at 10 million, a log–log slope of −0.503; the 95% interval covers the closed-form price in 94.65% of 11,000 seeded runs, and every test passes
Fig. 1.3 Call curve Black–Scholes call value against spot; K = 100, σ = 20%, r = 3%, T = 1. Schematic; the case study plots the computed values. Illustrative data
Role Author/Stack C++20, Python, PyTorch, pybind11, WebAssembly (Emscripten 6.0.13)/Date Jan 2026/Track ML, Quant
Problem. Learn to play a simplified multi-player Texas hold’em by self-play alone, with no poker knowledge given to the agents, and measure how they play.
Method. A C++20 game engine bound to Python with pybind11; Double DQN with one network playing every seat by self-play, at 100 big blinds with exact legal-action masking; nine candidates chosen between by a rule written before training; evaluation against fixed baseline bots on duplicate deals; the selected network re-implemented in C++ and compiled to WebAssembly for a browser demo.
Result. One self-play network (36,871 parameters) beats all three fixed baselines at 100 big blinds, on 40,000 fresh duplicate deals: +372 [345, 398] big blinds per 100 hands against random bots, +332 [314, 350] against calling bots and +18 [14, 23] against tight bots
Fig. 1.4 Self-play result by starting hand Mean result in big blinds per hand for each of the 169 starting hands when the selected agent plays all four seats against copies of itself, over 250,000 hands from deck seed 17000 with blinds and folded hands counted. From images/selfplay_result_by_starting_hand.png, made in the engine repository by its figures command from recorded seeds and copied unchanged; not re-run here.
(a) Systems: latency histogramIllustrative data
(b) Quant: Monte Carlo convergenceIllustrative data
Figure 1. One signature figure per track: (a) per-operation latency for the order book with p50 and p99 marked; (b) a Monte Carlo call price converging on the Black–Scholes value (S = K = 100, r = 5%, σ = 20%, T = 1) inside a 95% band. All panels are illustrative data: they show what each case study will report, not results.
Table 1. Archive: the 14 existing projects, grouped by track rather than A–Z. tbc marks a year still to be confirmed; the CVs or the commit history give years for 13 of these 14.
Year
Project
Track
Built with
Links
2024
Orderbook Simulation Order matching and VWAP in C++, a from-scratch neural network for price prediction, Python GUIs.
Discord Bot Music, word filtering, a guessing game and the obligatory ping–pong.
Software
Discord API
2022
Paddle-and-Ball Game on an FPGA A paddle-and-ball game in hand-written Verilog for a Nexys 4 DDR board: one button-controlled paddle against a computer paddle that follows the ball, drawn over VGA with a score for each side. Built as practice before an FPGA game coursework project that is not published.
Apr 2026 – PresentSoftware Engineer, Logos ResearchOwn and optimise the search service across more than 400,000 indexed items, using Python and C# to deliver fast, reliable retrieval at platform scale.
Aug 2025 – Sep 2025Freelance Software Engineer, Great Child NurseryBuilt the nursery website in React and TypeScript on Firebase, with client-side caching to cut database calls.
Jan 2024 – Jun 2025Junior Software Developer, Carbon & FinchDeveloped and optimised SQL jobs that analysed large volumes of sales-order data.
Jun 2022 – Sep 2022Junior Software Engineer, IDEX BiometricsDeveloped a MATLAB client for analysing fingerprint-sensing biometric systems.
Jun 2021 – May 2022Technical Director, Engineers Without Borders University of WarwickAutomated society operations with UiPath and C#, saving more than three hours each week.
Arcade: Snake, Block Stacker, Space Raiders, Coding NinjaProductionGame loops, collision and state transitions, with start, pause and restart in every game.
Order book demoLive demoThe C++ order book compiled to WebAssembly: place and cancel orders and watch them match, in the browser.
Poker demoLive demoThe hold’em engine and a trained agent compiled to WebAssembly: play hands against three copies of it, in the browser.
5 About and proof
I work across backend systems, search infrastructure, performance-focused development and quantitative finance projects. I trained as an electrical and electronic engineer at Warwick, which is where FPGAs and microcontrollers taught me to count cycles. Away from a keyboard: basketball, cycling, and more home automation than strictly necessary.