Researcher building applied tools at the intersection of epistemic infrastructure, deliberative democracy, AI systems design, and blockchain technology. MSc Data Science with Distinction (UWE Bristol, 2025). My work explores how we build the conditions for genuine reasoning, participation, and collective sensemaking through civic technology and public-facing, research-led software. Seven years in large-scale commercial platform operations gave me a practical understanding of how automated systems shape incentives, behaviour, and visibility: dynamics that increasingly structure epistemic and political life.
Project Summary: Can we identify common ground and characterise disagreement beneath rhetoric by mapping debate spaces and aggregating claims into canonical positions? The live project explores this through an experimental "deliberative debating" interface, with accompanying visualisations, report, and corpus.
Pipeline & Implementation: Designed and implemented a five-stage computational pipeline for mapping argument structure and rhetorical patterns in public debate: transcript ingestion, claim extraction, discourse skeletonisation, rhetorical technique classification, and canonical position aggregation. Analysed 17 debates and speeches, extracting 390 rhetorical techniques against an original 191-category taxonomy and classifying 456 claims from the corpus.
Methodology & Findings: Theoretically grounded in the Aristotelian distinction between dialectic and eristic argument, the project aims to foreground how arguments are made, and the underlying structure of debate, rather than whether conclusions are correct. The system identified 37 canonical positions across wealth tax discourse, including 5 points of cross-partisan agreement. Longer term, the project aims to explore whether identified points of contention can function as deliberative "springboards" toward more constructive public reasoning.
Project Summary: Can machine learning detect rhetorical manipulation techniques in news coverage of immigration, and can those detections be combined with bias and sentiment data into an interface that helps readers think critically rather than telling them what to think?
Architecture: End-to-end NLP pipeline using python scripts to collect + process 854 UK/US news articles — revived SemEval-2020 propaganda span detection in a containerised 2025 environment, enriched with VADER, TextBlob, spaCy NER, and bias ratings from Ad Fontes, MBFC, AllSides, and Ground News.
Front-end Design: Chart.js interface surfacing rhetorical manipulation patterns and sentiment.
Mission: Public-good platform connecting consumers to peer-reviewed evidence about food ingredients, with explicit transparency about funding sources and conflicts of interest behind scientific claims.
Technical: Collaborative React/Next.js MVP with barcode scanning (Quagga2), OpenFoodFacts API integration, ingredient risk lookup, and NOVA classification. PhD proposal in development at UWE Bristol.
Project Overview: Can DAO voting vulnerabilities such as whale swings and sybil attacks be mitigated using ML techniques to identify bad actors?
Design: Adaptive governance framework dynamically selecting between voting mechanisms based on participation distribution; Bootstrap-validated Isolation Forest anomaly detection for adversarial actors.
Impact: Applied to real on-chain data from ENS, Gitcoin, and SushiSwap; reduced winner-side Gini coefficient from 0.98 to 0.89.
Independent Consulting: End-to-end account management for multiple consumer brands across UK and EU: listings strategy, advertising, catalogue maintenance, compliance documentation, Brand Registry, and creative asset production.
Commercial Ownership: P&L responsibility for approximately £5m annual revenue across UK and EU: advertising strategy across multiple formats, promotional planning, pricing decisions, and cross-market forecasting.
Data & Systems: Built internal Excel dashboard consolidating cross-market performance; led full Seller-to-Vendor account migration; A/B testing and competitive analysis for listing optimisation.
Multi-Market Launch: End-to-end product launches across UK, DE, FR, IT, ES: content strategy, keyword research, advertising portfolio management, and asset delivery for new product lines.
Systems Design: Designed and implemented a product catalogue management system in Monday.com that remains in active company-wide use today.
Operations: Simultaneous multi-market catalogue management across 7 territories (UK, USA, DE, FR, IT, ES, NL); account health, compliance documentation, and full content lifecycle management.