Manchester Area, United Kingdom
Applied AI Engineer with 6+ years designing, leading and building production GenAI / ML systems across startups, enterprises and regulated industries. A strong combination of product thinking, frontier AI research, data science and engineering allows me to deeply understand problems and design solutions. Recognised as an "Innovation Engine" at Aviva for rapidly translating cutting-edge research into their first customer-facing AI agent. I specialise in: Evaluation systems: Golden datasets, LLM-as-Judge aligned with SME judgement, Recall / Precision / Coverage, synthetic data, root-cause analysis frameworks, observability platforms (Langfuse, Arize Phoenix) Knowledge structuring & ingestion pipelines: turning messy unstructured documents (legal, regulatory, technical) into AI-ready knowledge graphs and indexes - custom OCR, ontology design, agentic extraction, composable and versioned ingestion. Retrieval architecture: Embeddings, Vector Stores (pgvector, faiss) , Graph Databases (neo4j, falkordb), RAG, GraphRAG, hybrid retrieval, multi-stage pipelines, reranking, semantic routing. Agentic architectures: multi-stage agents, tool orchestration, frameworks (LangGraph, Pydantic AI), orchestration patterns, guardrails. Cloud deployments (AWS, GCP, Azure) or on-prem: FastAPI, ECS, Agentcore, Bedrock, Lambda, Docker, CI/CD, Vertex Agent Platform
Pragmaitic provide AI Strategy and Engineering to help startups and small businesses build reliable and accurate AI systems. With a focus on the context layer (Ingestion, retrieval and evaluation) around LLM systems.
Built the AI foundations (strategy, architecture and implementation) for Veridox's AI Fraud Detection System, with a strong focus on self-learning capability and increasing determinism. • Transitioned architecture from individual LLM calls to a composable modular agentic system • Established observability infrastructure whilst maintaining strict data-residency requirements • Set up evaluation framework and UI to test repeatability over key golden datasets • Established organisational configs to enable configurability of the system to client's key requirements whilst reducing cost and latency • Working directly with pre-sales clients to understand requirements, configuring and adapting the system to their needs and demonstrating solutions to support deal progression.
Led a team of three AI engineers for the Innovation Team that developed Aviva's first customer-facing AI agent. I established the technical direction and evaluation infrastructure that enabled rapid iteration and measurement against quality standards. I was recognised as an "Innovation Engine" for rapidly integrating cutting-edge approaches. • Advocated for and implemented GraphRAG approach when traditional RAG did not meet accuracy requirements. Designed ontology and knowledge graph architecture that delivered significant improvements in retrieval performance. • Built evaluation-first development infrastructure: collaborated with business stakeholders to capture customer personas and sample questions, generated diverse synthetic conversations covering edge cases and out-of-scope queries, implemented LLM-as-Judge aligned with internal QA processes. • Designed composable document ingestion pipeline enabling rapid experimentation with different extraction approaches. Established data architecture to version ingestion outputs and measure retriever performance (AI Flywheel). • Created root cause analysis framework combining qualitative feedback with data science metrics to diagnose whether issues stemmed from retrieval, generation, or knowledge base completeness. • Implemented a range of novel techniques and packages to enhance document ingestion (Docling, Langextract, Agentic Document Extraction), retrieval (hypothetical question embedding, graph traversals, reranking, pre and post filtering), agentic workflows (langgraph, semantic routing) and evaluations (Recall, Precision, Faithfulness, F1, Cohen's Kappa)
Conceptualised and developed a Gen AI solution enabling technical professionals in energy infrastructure to locate and utilise domain knowledge more efficiently. Solution deployed with innovation team at customer organisation. • Implemented GraphRAG solution for efficient technical information retrieval across complex regulatory documentation. • Led product demonstrations to technical and non-technical stakeholders, translating complex AI capabilities into business value propositions. • Developed custom document processor for legislation from multiple providers where no suitable off-the-shelf solution existed. Built OCR pipeline capable of identifying paragraph boundaries, clause numbers, and document structure to transform unstructured legal text into AI-ready knowledge graphs. • Set up evaluations and monitoring (Langfuse) as well as CI/CD (LLMOps)
Built self-invoicing automation system replacing manual procurement processes. Developed dashboard enabling staff to monitor and send invoices automatically, saving an estimated £100k/year. Developed forecasting models to predict future energy prices. Used when negotiating Power Purchase Agreement (PPA) deals with electricity providers.