Projects

Active

AlphaSignal Research Agent

Active
LangGraphPythonFAISSRAGAnthropicvectorbt

Multi-agent pipeline that reads SEC EDGAR 10-K and 10-Q filings and generates structured investment signals. A LangGraph graph orchestrates three agents: a RAG researcher (FAISS + MMR retrieval over filing chunks), a Claude-powered signal extractor producing cited, schema-validated directional calls, and a vectorbt backtester running a 2-year trailing window with benchmark comparison. Includes LangSmith evaluators for citation grounding, confidence calibration, and RAG faithfulness.

LifeOS

Active
FastAPIReactPythonMonte Carlo

Personal finance calculator covering the things UK 20-somethings actually care about: deposits, savings timelines, student loans, and investment targets.

Chase UK FAQ RAG System

Active
PythonHNSWLibAzure OpenAILitestarGraphQL

Production RAG pipeline for Chase UK's mobile banking FAQ assistant. Async Python monorepo with LLM-driven re-ranking, multi-layer safety guardrails, and React 19 frontend. Deployed on AWS EKS via Terraform.

Past

2D→3D Architectural Style Transfer

TensorFlowVGG-19MiDaSPython

MEng dissertation project. Applied neural style transfer to 3D architectural models using depth estimation and mesh projection, translating 2D architectural styles onto 3D geometry. Published in the Bartlett Show 2022 Book.

Read dissertation →

Pit Signal Accuracy — Alpine F1

StatisticsError MinimisationPython

Statistical modelling project improving pit lane signal accuracy by 91% using error-function minimisation. Results adopted directly by the Strategy and Race Track teams.

Process Mining — Alpine F1

Process Mining3DXPython

Analysed 3DX engineering logs to identify 80% of process divergences across three model years. Findings adopted by Enterprise Architecture and PLM teams.