AI Architecture & Strategy
End-to-end architecture for LLM, agentic and ML systems. Roadmaps that survive contact with production.
[ 03 / CONSULTING ]II · For companies shipping AI into production
I take on a small number of consulting engagements with software companies and product teams building production AI. From architecture and strategy down to shipping - banks, ed-tech, recruitment, OCR, generative AI.
[ ENGAGEMENTS I TAKE ON ]
RAG & Agentic Systems
RAG that doesn't hallucinate. Agents that do real work. Evaluation harnesses you can trust.
Computer Vision & OCR
Document AI, OCR pipelines, vision models from prototype to scaled inference.
MLOps & Inference Optimization
Training, retraining and serving pipelines across cloud and on-prem - built for cost and latency.
[ HOW I WORK ]
Three ways to start.
Architecture Review
2 - 3 weeks
1 - 2 new engagements per quarter
A focused audit of your AI architecture, data and evaluation loops. You receive a written assessment and a prioritised roadmap your team can execute without me.
Best for: Teams with an AI system that underperforms - or a plan that needs a second opinion before the build starts.
Build Engagement
6 - 12 weeks
1 - 2 new engagements per quarter
Hands-on delivery alongside your team - from architecture through a shipped, measured system. I design the evaluation loops and leave behind something your engineers can operate.
Best for: Product teams shipping their first serious AI feature into production.
Fractional AI Lead
Quarterly, renewable
1 - 2 new engagements per quarter
Embedded with your team a few days a week, owning AI direction - architecture, hiring input, delivery and review. Senior AI leadership without the full-time hire.
Best for: Companies that need AI direction before a full-time head of AI makes sense.
Engagements are scoped individually. Rates shared on the first call.
[ SELECTED WORK ]
Challenge
Digital banking transaction volumes were outgrowing rule-based fraud monitoring. The bank needed risk decisions in real time, with reasoning an analyst could act on.
Approach
Architected an AI-driven fraud detection platform - real-time risk scoring, behavioural analytics and LLM decision support.
Graph MLReal-time scoringLLM decision supportRisk analytics
Challenge
Grading printed and handwritten student work by hand does not scale. Fonix needed recognition accurate enough to trust and cheap enough to run at volume.
Approach
Led R&D for an AI grading system - recognition for printed and handwritten documents, synthetic data generation, scalable inference.
OCRHandwriting recognitionSynthetic dataScaled inference
Challenge
Screening high volumes of CVs manually was slow and inconsistent. OXO needed parsing and ranking reliable enough to put in front of recruiters.
Approach
End-to-end architecture for an LLM-powered recruitment automation platform - CV parsing, ranking, fine-tuning and evaluation.
LLM fine-tuningCV parsingRankingEvaluation harness
Challenge
Call-centre QA relied on manual sampling of a small fraction of interactions. Digital365 needed automated coverage across its document and voice workflows.
Approach
Enhanced OCR systems and architected AI-powered QA solutions for call-centre operations.
OCRQuality assuranceSpeechAutomation
[ CLIENT ROSTER ]
2026 - Present
Lead AI Consultant
Enhancing OCR systems and architecting AI-powered QA solutions for call-center operations.
OCRQA automationCall-centre AI
2025 - 2026
Lead AI Consultant
Leading R&D for an AI grading system - recognition for printed & handwritten documents, synthetic data, scalable inference.
OCRHandwritingSynthetic dataInference
2025 - 2026
Lead Consultant · Data Science
Architecting an AI-driven fraud detection platform - real-time risk scoring, behavioural analytics, LLM decision support.
Fraud detectionGraph MLRisk scoringLLM
2025
Lead Consultant · Generative AI
End-to-end architecture for an LLM-powered recruitment automation platform - CV parsing, ranking, fine-tuning, evaluation.
LLMFine-tuningCV parsingEvaluation
2021
Machine Learning Consultant
Applied ML consulting across product workstreams.
Applied ML
2025
AI Consultant
Applied AI consulting across product and automation workstreams.
Applied AILLM systemsAutomation
[ BOOK A CONSULT ]
Have an AI project that needs to actually ship?
I usually take on one or two new engagements per quarter. If you're scoping production AI work - architecture review, RAG, agents, MLOps, fraud detection, document AI - tell me what you're building.