About Me
Data scientist and ML engineer with fifteen years of software engineering experience spanning aerospace systems, formal verification, custom silicon design, and production AI infrastructure β translating rigorous analytical methods into deployable, measurable solutions.
My theoretical foundations include active contributions to mechanised proof development in Lean 4 and Mathlib covering real analysis, combinatorics, linear algebra, and propositional logic, extended into domain-specific formalisation: an Aircraft Parts Volatility Index model, war economy market propagation dynamics, and formal aerospace pricing guardrails. Core stack: JAX, PyTorch, Polars, scikit-learn, FastAPI, and LangChain with LangGraph on AWS and Railway.
My Neo4j Aura hackathon entry Shindo (ιεΊ¦) is a Japanese earthquake impact simulator β a knowledge graph across ~33,000 nodes enriched with a thirteen-class OWL ontology, Lettria semantic extraction, GEBCO bathymetric data, NOAA tsunami records, USGS seismic data, and a Claude tool use agent generating structured damage assessments at query time. My Systolic BNN Accelerator was separately selected for production on SkyWater 130nm silicon via Tiny Tapeout SKY26b β my second custom silicon tapeout.
π» Skills & Tech Stack
π Projects
ιεΊ¦ Shindo β Japan Seismic Risk Graph
Cascading seismic risk intelligence graph connecting 20,000 earthquakes, fault zones, tsunamis, nuclear facilities and prefectures. Neo4j Aura Agent hackathon entry.
View ProjectSystolic BNN Accelerator (V2) β Selected for Fabrication
8-neuron XNOR-popcount accelerator on a 1Γ1 Tiny Tapeout tile. Systolic dataflow, signed bias, balanced popcount tree. Selected for fabrication from a competitive group submission.
View ProjectAMP SPC Project
ML-driven SPC for 1590nm amp splice testing. 21,420 tests analysed, 99.7% yield confirmed, Β£143K CONQ identified. SVR production model.
View ProjectMachine Learning MCMC Model
XGBoost + NUTS hierarchical Bayesian regression for aircraft parts pricing. MAE $4,238 with full posterior predictive intervals.
View ProjectRisk Algorithm β Aircraft Parts
GARCH volatility modelling and MCMC inference with Lean 4 formal proofs β Banach fixed-point theorem applied to a live commercial pricing system.
View ProjectLean Theorem Proving
Over 30 machine-verified proofs in Lean 4 and Mathlib spanning real analysis, topology, algebra, combinatorics, and war economy market propagation.
View Project16-Neuron Binary Neural Network
BNN inference layer in silicon β 16 perceptrons classifying an 8-bit input simultaneously in a single clock cycle. No CPU, no software.
View ProjectBioinformatics β Squamous Cell Carcinoma
Computational analysis using Python and Jupyter, surfacing statistically significant patterns in genomic and clinical data.
View Projectπ Work Experience
AJW Aerospace β Data Scientist
Architected a dual-model production pricing pipeline combining XGBoost gradient-boosted regression and MCMC NUTS hierarchical Bayesian inference via NumPyro β transforming the dataset median price-to-bracket ratio from 9% to 101% through a critical data quality filter, enabling bracket-reasonable predictions across 184 unique part numbers for the first time
Built GARCH volatility modelling pipeline in JAX for aerospace parts risk assessment, with core mathematical properties formally verified in Lean 4 and Mathlib β covering pricing guardrails, volatility spike detection, and market propagation via the Banach fixed-point theorem, providing machine-checked correctness on live commercial decisions
Designed and delivered autonomous agentic AI systems consistently within 8β12 week cycles, achieving ROI improvements of 15β30% above projections; applied Monte Carlo simulation, DCF analysis, and decision theory frameworks for strategic risk assessment
Engineered end-to-end ML infrastructure on AWS SageMaker, Lambda, and EMR alongside Google Cloud Vertex AI and BigQuery, reducing deployment time for new models from weeks to days through reusable pipeline components
ASN Submarine Cables β Data Scientist & ML Engineer
Converted corrective to predictive maintenance across fibre optic and laser manufacturing cells using ML failure prediction and machine vision anomaly detection β reducing time-to-detection of process drift from days to real-time and measurably decreasing unplanned downtime
Implemented SPC across 1590nm amp splice testing β identifying out-of-control events across over 21,000 tests and surfacing a significant Cost of Non-Quality directly attributable to blind splicing; recommended and validated corrective actions that addressed the root cause
Applied MCMC predictive models and statistical process control to reduce defect escape rates; led Kaizen events and Lean Manufacturing initiatives improving cell DPMO KPIs through DMAIC and 8D root cause elimination
Designed 3D-printed manufacturing fixtures in Fusion 360 and Creo, eliminating external procurement lead times and reducing component delivery from weeks to same-day production
McLaren Racing β Complex Data Analysis
Implemented Statistical Process Control and predictive ML models across vehicle production lines in Python, MATLAB, and Octave β reducing manufacturing defect rates and improving first-pass yield on high-tolerance components
Led FMEA and PPAP protocols identifying and eliminating failure modes before production release; applied A/B testing frameworks to validate process improvements in cycle time and defect rates with statistical rigour
Delivered Power BI dashboards for real-time production monitoring, enabling data-driven decision-making at line and management level and reducing the time from data collection to actionable insight
East Surrey College β Aeronautical Engineering Lecturer
Delivered mathematics, physics, Python programming, CAD, and Six Sigma DMAIC to aerospace engineering students; designed hands-on VSM and SPC exercises that connected theoretical quality methods to real manufacturing contexts