Academic Projects @ Warwick

• 

  Multi-Agent Car Parking using Reinforcement Learning

  Undergraduate thesis applying reinforcement learning to simulate and control groups of autonomous vehicles using PPO and Unity ML-Agents.

  • Trained agents with IPPO and MAPPO, and implemented DQN from scratch, using TensorFlow/Python.
  • Designed a flexible environment modelled as an MDP with independent agents and dynamic goals, implemented in Unity using C#.
  • Achieved a 98.1% parking success rate with up to 7 cars, outperforming single-agent baselines.
  • Explored collaborative and competitive behaviours with variable communication and density.
  • Revealed novel group dynamics such as 'leaky' collaboration and competition-induced cooperation.
  • Scaled training on the university's HPC using Slurm.
  • Presented at ICUR '22.
  • Awarded 85% for the thesis (top 2% / 250 in cohort).
  PythonC#UnitySlurmTensorFlow
  View CodeRead on arXiv
• 

  MiniC Compiler

  Implemented a compiler frontend for a simple C-like language using modern C++17 and LLVM.

  • Developed lexical analysis and parsing with Flex and Bison.
  • Generated LLVM IR through syntax tree traversal.
  • Implemented semantic analysis and robust error reporting.
  • Awarded first-class honours for the project.
  C++LLVM
  View CodeRead Report
• 

  Physics Simulation Optimisation

  Optimised a two-dimensional computational fluid dynamics simulator.

  • Implemented low-level optimisation techniques on the C code, making it run in 1/10th of the original time (~80s to ~7s).
  • Parallelised via multi-threading with OpenMP and vectorisation with SSE intrinsics, after removing inter-loop dependencies.
  • Applied general optimisation techniques such as loop unrolling, loop fusion & fission, precomputing static conditionals, and removing unexecuted code.
  COpenMP
• 

  Gaussian Mixture Model Clustering with EM Algorithm

  Implemented clustering on the Iris dataset using Gaussian Mixture Models and the EM algorithm, improving accuracy over K-Means.

  • Achieved 98% max cluster assignment accuracy after 19 EM iterations.
  • Used PCA for dimensionality reduction and data visualisation.
  • Developed and analysed the E-step, M-step, and label assignment for EM clustering.
  Python

• Packet Sniffer

  Implemented a packet sniffer to detect SYN Flood, ARP Poisoning and URL Blacklist attacks.

  • Low-level C implementation of IP and TCP packet header parsing.
  • Multi-threaded implementation to handle high load of incoming packets.
  • Used a work queue and mutex locks to handle high loads and prevent race conditions.
  C
• 

  Timetable Scheduler

  Implemented a class timetable scheduler in Python.

  • Abstracted the problem into a game of 'Tetris'.
  • Modelled scheduling around time and cost constraints as a Constraint Satisfaction Problem (CSP), and implemented CSP solving algorithms.
  • Found the optimal assignment by implementing informed search algorithms.
  Python
• 

  TurtleBot Planning, Localisation and Control

  Implemented motion planning, control, sensing, and localisation using ROS and Turtlebot3 in the Gazebo simulator.

  • Implemented in Python using ROS.
  • Developed square path navigation using velocity control and PD controllers.
  • Implemented closed- and open-loop obstacle avoidance with laser scan data.
  • Applied particle-filter-based localisation with covariance analysis at waypoints.
  • Designed wall-following and dynamic adaptation behaviours.
  PythonROS
• 

  Pong

  Implemented the classic 'Pong' game on the screen of an oscilloscope via an ARM controller.

  • Implemented core game logic, including game loops and rendering.
  • Built an API between the oscilloscope screen and the C code, providing primitive operations (drawing rectangles and other shapes) later combined to form the whole game.
  CARM
• 

  Derivative Trade Manager

  Software Engineering group project sponsored by Deutsche Bank, involving the design and implementation of a full-stack application to manage derivative trades.

  • Delivered the project through the full Software Engineering lifecycle (requirements gathering, design, implementation) in a group of 6.
  • Backend implemented in Java with an SQL database, frontend built with React (JavaScript), communicating via a REST API.
  • Automatically learns, detects and corrects input errors using a Naive Bayes model.
  • Scalable implementation via pagination and query indexing.
  JavaJavaScriptSQL
• 

  Witter

  Implemented the backend data store for a 'Witter' application (akin to Twitter).

  • Implemented hash tables, search trees, and other data structures from scratch.
  • Balanced time and space complexity trade-offs to optimise for scalability.
  Java
• 

  Scratch Clone

  Implemented an interpreter that computes the resulting memory state from an input program in Scratch block format and the initial memory state.

  • Implementation used a Monad Transformer and other high-level abstractions.
  • Obtained full marks for the implementation, and awarded two Purple λ awards for achieving marks among the year's highest for Functional Programming.
  Haskell

• PLM Parser and Evaluator

  Implemented a parser and evaluator for a made-up programming language (Programming Language of the Moment).

  • Translated the rules of the language into a Formal Grammar (functions, parameters, variables, basic mathematical operators and positive integers).
  • Implemented a parser using JavaCC and an evaluator using Java.
  • Obtained full marks for the implementation.
  JavaJavaCC

• Propositional Logic Prover

  Implemented a logical evaluator to determine whether an input theorem is a tautology.

  • Implemented the resolution algorithm for propositional logic formulas to determine tautology, in a divide-and-conquer fashion.
  • Used Prolog for the implementation, making it entirely functional.
  Prolog