I build AI, data and workflow products for complex operational environments.
I am currently Principal Product Manager – OS at Sedna, where I shape the reusable workflow and execution layer of the operating system for global shipping. Before moving into product leadership, I served as Head of AI, ML & Data and Lead AI Architect, leading production AI, data architecture and customer-facing automation.
My work spans hands-on Python and cloud engineering, product strategy, technical leadership and company building. The common thread is turning unstructured data, fragmented systems and manual workflows into production tools that people can depend on.
Founder → acquisition · AI and data leadership · Product strategy · Production systems
Flagship work
Flytta — AI workflow automation for global trade
Customs and trade teams receive critical information through PDFs, spreadsheets, emails and scanned documents, then repeatedly re-enter that information into operational systems.
I led the technical and product development of Flytta's commercial platform. It ingested documents through email and APIs, extracted structured data using OCR and AI, validated and enriched that data, applied customs-specific business rules and generated outputs for declaration systems.
My role covered the entire product lifecycle: problem discovery, architecture, hands-on Python development, customer implementation, enterprise integration, product positioning and commercialisation.
Impact: Flytta was deployed into live, high-volume customs workflows, reducing some multi-hour manual processes to seconds. The company was acquired by Sedna in 2025, with its customs automation capabilities becoming part of Sedna AI.
Built with: Python, AWS, PostgreSQL, OCR and document AI, LLMs, APIs, event-driven pipelines and enterprise customs integrations.
Read the acquisition announcement
Sedna — building the execution layer for the operating system of shipping
I joined Sedna after it acquired Flytta in March 2025. I began as Lead AI Architect, working directly with enterprise customers as a forward-deployed engineer, before taking responsibility for Sedna's global AI, machine-learning and data function.
I now work in Sedna's OS product pillar as Principal Product Manager. My current focus is the workflow and execution layer: understanding how shipping teams actually operate, identifying patterns that repeat across customers, and turning those patterns into reusable platform capabilities rather than isolated integrations.
I own the product direction and backlog for this area, working with an engineering squad from customer discovery through technical definition, delivery, adoption and measurement.
Selected work at Sedna includes:
- MCP and agent infrastructure: I designed and helped productionise Sedna's Model Context Protocol capability, connecting AI assistants and agents securely to operational shipping data. The serverless implementation was built around AWS and FastMCP, with the tool surface deliberately reduced from 25 tools to 15 to improve agent performance and usability. Production concerns included authentication, tenant separation, permissions, observability and reliable tool design.
- Enterprise AI workflows: I worked alongside major maritime and logistics customers to turn high-volume communications into structured data and executable workflows. This included LNG noon-report analysis against voyage instructions, customs import and export automation, intelligent extraction, tagging and operational anomaly detection.
- AI and data leadership: As Head of AI, ML & Data, I led a distributed team of AI software engineers, set the technical direction for production AI and owned major parts of Sedna's data architecture, including SageMaker, Snowflake, data pipelines, governance, MLOps and model-lifecycle standards.
- Productisation: In the OS role, I am translating customer-specific discoveries into reusable workflow primitives, actions and automation patterns that can operate across customers and use cases. This connects customer need, product strategy, data architecture, APIs and execution reliability in one product area.
Sedna's MCP capability allows supported assistants to work with messages, contacts, users, job references, categories and team inbox data. It has been adopted by enterprise customers to interrogate their own operational information and deploy agents that use shipping data within their existing workflows.
I have also represented Sedna externally, including an AWS Let's Build a Startup session on production MCP servers and a presentation at AWS London GenAI & Data Day on using MCP to accelerate customer onboarding. I was selected for the Digital Leaders AI 100 cohort, recognising people applying AI to practical problems across the UK.
This progression—from founder and engineer, through AI and data leadership, into principal product management—means I can move between the customer workflow, commercial case, product roadmap, data model, API boundary, model behaviour and production architecture.
Explore Sedna AI · Read AWS's production MCP write-up
Seer — building a data science company
I co-founded Seer as a data science lab focused on applying machine learning, AI and data engineering to industries with difficult operational problems.
I helped take the company from early consulting engagements to a multidisciplinary team delivering production systems and developing its own products. My work combined sales, discovery, product management, cloud architecture, engineering, hiring, partnerships and grant-funded innovation.
Selected outcomes included:
- Developing and commercialising Flytta before its acquisition by Sedna.
- Delivering production systems across logistics, maritime operations, autonomous vessels, customs and business intelligence.
- Securing and delivering projects supported by Innovate UK and Digital Catapult.
- Building relationships across industry, government, academia and the UK technology ecosystem.
- Turning one-off client problems into reusable technical capabilities and products.
Running Seer taught me how to move between commercial strategy and technical detail without treating them as separate disciplines.
NCB Hazcheck / Exis Technologies — AI for dangerous-goods safety
NCB Hazcheck provides dangerous-goods compliance, validation and cargo-screening systems for maritime transport. The wider Hazcheck product family is used by nine of the world's ten largest container lines, placing the work in an operational environment where accuracy, reliability and explainability have direct safety consequences.
Through Seer, I led work with Exis Technologies to introduce advanced data, analytics and machine-learning capabilities into its hazardous-cargo monitoring platform.
We developed custom analytics infrastructure and AI models over a dataset containing more than 31 million records. The work supported hazardous-cargo monitoring, rule recommendation and near-real-time visibility of cargo moving through the platform.
The client-facing models used natural-language-processing techniques including tokenisation and generative indexing to identify patterns, connect cargo information to relevant rules and surface useful information from complex maritime safety data.
My contribution covered customer discovery, solution architecture, technical direction and the translation of specialist dangerous-goods knowledge into deployable data products. We worked closely with Exis Technologies' own development team so that the new data capabilities could be integrated into services used across global container-line operations.
Impact: The project gave Hazcheck deeper insight across a large and operationally important dataset, improved internal and customer transparency, supported near-real-time cargo monitoring and created a scalable foundation for further AI-assisted safety capabilities.
Built with: Python, machine learning, NLP, large-scale data processing, custom analytics infrastructure, APIs and cloud services.
COG Legal — machine learning for legal invoice and spend intelligence
COG Legal helps in-house legal and insurance teams understand and control their external legal spend.
Through Seer, I led data-science and product work that introduced natural-language-processing-based machine learning into COG Legal's existing legal-spend platform. The objective was to increase the volume of documents the organisation could process while preserving the specialist judgement of its legal-spend experts.
The system processed legal invoices and supporting information, classified legal billing activity, identified relevant patterns and exceptions, and transformed unstructured billing narratives into data that could be analysed consistently.
Rather than attempting to replace expert review, we designed the capability as a human-in-the-loop system. Machine learning handled repetitive classification and detection work, while legal-spend specialists retained responsibility for nuanced decisions, challenges and client recommendations.
The work operated at two levels:
- Reusable platform capability: NLP-driven document processing, billing classification, detection and structured-data generation that could be applied across multiple customers.
- Customer-specific implementation: adapting data pipelines, classification logic, dashboards and reporting to different billing guidelines, law-firm panels, tax structures, currencies, regions and client operating models.
The platform supported customer programmes across the public sector, automotive industry and international markets.
Selected customer outcomes reported by COG Legal include:
- A large English local authority using Legal eTeam identified £1.4 million in savings and recovered time equivalent to one full-time employee.
- A major automotive customer achieved more than £5 million in savings, used spend data to support a global law-firm panel process and was able to move senior lawyers away from billing administration and onto more strategic work.
These results came from COG Legal's wider service—combining expert legal invoice review, customer processes and its digital platform—rather than from an ML model acting alone.
Impact: The technology allowed COG Legal to process legal documents at greater scale, reduce repetitive classification work, provide clearer spend information and support new customers across different industries and regions.
Built with: Python, NLP, supervised machine learning, document-processing pipelines, data classification, cloud services and analytics dashboards.
Explore COG Legal's AI platform · Read the public-sector customer case study
Oshen — mission control for autonomous ocean robots
Oshen develops small autonomous vessels designed to collect data from the world's oceans.
I helped design and build the cloud platform used to manage these vessels as a fleet. The system covered mission configuration, vessel monitoring, telemetry ingestion, operational status and the presentation of collected data.
The central product challenge was not simply displaying vessel locations. It was designing an operational system that allowed users to understand what an autonomous fleet was doing, identify problems and manage missions remotely.
My contribution included: platform architecture, data modelling, cloud pipelines, API design, product discovery and translating operational requirements into a usable interface.
Built with: Python, cloud services, event-driven data pipelines, APIs, geospatial data and vessel telemetry.
Teesside Freeport — inventory and customs movement control
I helped design and deliver an inventory-control system for goods moving through a freeport customs environment.
The platform connected physical goods movements with customs records, inventory positions and operational notifications. It captured information including quantities, weights, values, declaration references, duty and VAT status, transport details and movement history.
A major requirement was auditability. Every movement needed to be traceable while still allowing operational teams to work quickly with incomplete or changing information.
My contribution included: product discovery, data architecture, API and workflow design, customs-domain logic, cloud implementation and operational reporting.
The project brought together several areas that have shaped my career: complex regulation, fragmented data, physical operations and the need for dependable software.
Technical research and experiments
Quantum machine learning for late-delivery prediction
As part of Digital Catapult's Quantum Technology Access Programme, I developed a hybrid quantum-classical classification use case for predicting whether a delivery would arrive late.
The work involved translating an ordinary industrial machine-learning problem into a form suitable for experimental quantum hardware, using ORCA Computing's photonic quantum technology and associated development tools.
The most valuable part of the project was learning how to evaluate emerging technology honestly: understanding what quantum methods could demonstrate, where classical models remained stronger and what would need to change before a commercial advantage became possible.
Read Digital Catapult's programme write-up
Football transfer and career modelling
I built a machine-learning project to model football-player development, future transfers and market value.
The project combined data from multiple football datasets, including BeSoccer, SciSports and Transfermarkt. The difficult part was not initially the model: it was resolving players across inconsistent sources, constructing reliable historical records and engineering features that represented career trajectory.
I then experimented with models for predicting future clubs, transfer behaviour and changes in player valuation.
This was an early example of a theme that has followed me through most of my work: the quality of the data model and pipeline usually matters more than the novelty of the final algorithm.
Other selected projects
Automated carbon measurement from customs data
A system for calculating shipment emissions from trade and declaration information, developed through transport innovation work.
Rune — natural-language data analysis
An AI-powered analytics concept that allowed users to query organisational data conversationally, generate charts and surface trends without needing to write SQL.
Educational document intelligence
Machine-learning and document-processing work covering handwritten material, structured assessment and automated data capture.
Invoice and document validation
Operational pipelines that extracted, compared and validated information across invoices, spreadsheets, emails and internal systems.
Maritime and logistics analytics
Data platforms covering vessel activity, cargo operations, operational performance, movement tracking and business intelligence.
How I work
Start with the real workflow.
I spend time understanding what people actually do, including the spreadsheets, workarounds and exceptions that rarely appear in a formal specification.
Build end to end.
I am comfortable moving from product discovery and architecture into Python, data models, cloud infrastructure, APIs, deployment and customer implementation.
Treat AI as a system, not a feature.
A useful AI product needs reliable data, evaluation, workflow design, fallbacks, monitoring and a clear understanding of where human judgement belongs.
Connect technical and commercial decisions.
Architecture, positioning, implementation cost and customer value are closely related. My strongest work has come from considering them together.
Current interests
I am currently most interested in:
- AI-native enterprise products and operating systems.
- Agentic workflows that can reliably complete multi-step work.
- Fine-tuning and model adaptation for specialised tasks.
- Data and platform primitives that make enterprise AI dependable.
- Applying modern AI techniques to established, operational industries.
- Building useful personal AI systems for programming, travel, training and everyday work.
Writing and open work
I write about applied AI, data products, global trade, technology and the process of building things.
Read my writing · Explore my GitHub
Contact
I am most useful where a problem needs product judgement, technical depth and someone willing to get hands-on.