AI-powered platforms that turn decades of industrial data into decisive operational intelligence.
Industrial data science, end to end. From legacy archives to live operations, we turn raw data into AI systems, analytics and intelligence the people who run your assets can actually use.
End-to-end intelligence systems, from raw legacy documents to production-grade applications that engineers actually use.
Retrieval-Augmented Generation pipelines grounded in your own inspection records, drawings and operational documents. Production LLM integration, vector databases, document intelligence.
Architecting and shipping the application layer — combining LLM reasoning, vector search and industrial data into platforms designed for the workflows of integrity engineers and asset managers.
Turning legacy inspection reports, RBI studies and unstructured operational documentation into clean, queryable, analysis-ready datasets — at scale, with AWS Lambda, Textract and bespoke pipelines.
Automated interpretation of technical drawings, isometrics and P&IDs — CNN and Siamese network architectures that classify, extract and digitise the assets buried in your archive.
Decision systems that schedule under real-world constraints — crew skills, bed limits, materials, dependencies, area access. Solver-backed plans that maximise value, re-optimise on disruption, and explain themselves to the people holding the budget.
Time-series projection and self-correcting models calibrated against measured ground truth. Earned-value forecasting, demand projection, equipment-life prediction — built with two-model designs that sidestep the well-known fragility of single-equation forecasts.
A selection of platforms designed and delivered to clients across asset-intensive industries — oil & gas, renewables, industrial operations. Each shipped as a working application that engineers use in their daily work.
An AI-powered platform that reorients industrial data around the equipment itself — letting engineers interrogate decades of inspection records, RBI studies and operational documentation in natural language, with traceable, source-linked answers.
A vector database and Retrieval-Augmented Generation platform purpose-built for industrial document intelligence. Search, synthesise and reason over the technical archive — drawings, reports, specifications — with citations engineers can trust.
Modular ETL software that ingests unstructured legacy documentation — scanned reports, technical drawings, multi-decade inspection histories — and outputs structured, analysis-ready data. Deployed across multiple client engagements.
Automated interpretation of piping isometrics, P&IDs and technical drawings using convolutional and Siamese neural networks — converting hand-drawn and scanned engineering documents into machine-readable, queryable assets.
Full-stack platform replacing spreadsheet-based campaign planning with a CP-SAT constraint optimiser — maximising included work-item value subject to crew skills, POB bed limits, materials and dependencies. Break-in impact analysis pins the existing schedule so users see only the genuine knock-on, not solver churn.
The personal lab — where the same engineering discipline gets applied to a problem I actually own. A test bed for the local-first patterns that show up in client work.
Household-scale, multi-user health and fitness tracking — built on local infrastructure. Smart-scale ingestion, on-device OCR + LLM nutrition parsing (no third-party calls), Apple Watch GPX rendering on Leaflet, and a self-correcting weight prediction model. PWA-installable, deployable from a PC to a Raspberry Pi.
The components that typically ship in a Data Rich application — an intake card, an acceleration curve, a risk surface, a throughput meter, a grounded answer, an anomaly view, and an attribution layer. Figures below are illustrative.
A Data Rich application typically ships in three layers — a data substrate, the intelligence services that reason over it, and the application surface engineers actually use.
"Show me the corrosion history of caisson C-4 over the last decade — and flag anything outside the inspection envelope."
Caisson C-4 exhibits localised pitting in the splash zone, first noted INSP-2019-04. Subsequent ROV surveys (2021, 2023) confirm no propagation beyond the original wear pad. One anomaly — wall-thickness reading below tolerance — flagged in RBI-2024-08; investigate before next campaign.
Lead Data Scientist with a First Class Honours degree in Data Science and over five years building AI-powered data solutions for asset-intensive industries — from document controller at Lockheed Martin and information management at Spirit Energy, to designing and delivering the QUERY platform.
Specialist in LLM integration, vector databases, full-stack application development and the digitisation of decades-old industrial archives. Track record of managing technical delivery and client engagement across oil & gas, renewables and industrial operations.
— Data, distilled.
If your organisation is sitting on decades of unstructured technical data — and the people who understand it are retiring — we should talk.
richard@data-rich.co.uk