Work

Ongoing MEng thesis at Imperial College London. Developing MobileUNet for real-time fetal heart segmentation, AutoFHR for automated heart rate estimation, and L-FUSION for landmark-guided probe navigation. Targeting deployment on portable ultrasound devices for low-resource settings.

Developed production voice synthesis at DeepSearch Labs using MaskCycleGAN-TTS for voice conversion and VITS for end-to-end speech synthesis. Enabled personalised ad audio generation at scale.

Implemented GPT-2 (124M parameters) from scratch as part of Eureka Labs. Built the full transformer decoder architecture including multi-head self-attention, positional encoding, and byte-pair encoding tokenizer. Trained on OpenWebText.

Research project benchmarking large language models across 70+ languages on diverse NLP tasks. Evaluated multilingual performance gaps and contributed to understanding of cross-lingual transfer in generative AI.

Built a cryptocurrency pairs trading system using cointegration analysis, Kalman filters for dynamic hedge ratios, and mean-reversion signals. Backtested across multiple exchange pairs with transaction cost modelling.

Competed in Optiver's trading academy, developing options market-making strategies with risk-adjusted position management. Achieved 3rd place out of 100 participants through dynamic delta hedging and volatility modelling.

Built a teaching operating system in C at Imperial, implementing thread scheduling, user programs, virtual memory, and a file system. Scored 95%, placing in the top 5 out of 200 students.

Won a national AI hackathon by building an automated trading system using AI agents. Designed multi-agent coordination for market analysis, signal generation, and trade execution.

Secured 1st position in Morgan Stanley's arbitrage challenge, competing across manual trading and quantitative reasoning rounds. Demonstrated strong market intuition and analytical problem-solving under pressure.