Enterprise AI

What we build.

Six practice areas and around thirty capabilities, delivered by engineers and scientists who ship alongside your team. Four of the technologies below are our own, taken from research through to deployment.

Practice areas

Depth over breadth.

We go deep on every problem rather than hand across a cookie-cutter solution. Each area below is staffed by practitioners who write the code and design the systems.

AI & Automation

Intelligent systems that reach production, from agents and language models through to computer vision and process automation.

Our own AI products →

Agentic AI: AI agents, tool use and RAG, multi-agent systems
Machine learning: Deep learning and reinforcement learning, for classification, prediction and recommendation
Natural language processing: Conversational AI and text analytics
LLM tuning and serving: Fine-tuning, optimisation and serving at low latency and high throughput
Human data evaluation: RLHF and DPO, red-teaming, hallucination detection and reward model data
Computer vision: Recognition, object detection and visual inspection
Intelligent automation: AI-driven decisioning and process orchestration
AI for gaming: Intelligent NPCs, procedural content and player analytics

Engineering & Cloud

Mission-critical systems: the platforms, the pipelines and the hardware underneath them.

Hardware and system design →

Infrastructure modernisation: Cloud migration and microservices, out of legacy estates
Cloud analytics: Data warehousing and cloud-native analytics platforms
Cybersecurity: Threat identification and enterprise-grade protection
DevOps and platform: Full-cycle DevOps across the delivery lifecycle
Embedded systems: Custom firmware, IoT gateways, edge and wearable computing
Cloud platforms: Landing zones, migration and operation for regulated and data-resident workloads
Cloud rendering: Rendering infrastructure for visual effects, animation and simulation

Product & Design

Strategic guidance and full-stack engineering, from first prototype to shipped product.

Prototyping and MVP →

Product engineering: Full-stack development and user experience
PoC and MVP development: Rapid prototyping, feasibility analysis and investor-ready demonstrations
UI and UX design: Research-driven, accessible interface design
Web and mobile applications: Cross-platform builds for performance and scale

Data & Forensics

Getting value out of data, and standing behind it when it is examined.

Enterprise software →

Data engineering: Data strategy, pipelines and platforms
Digital forensics: Investigation and analysis of digital evidence
Scientific computing: High-performance simulation, modelling and numerical analysis
Optimisation: Operations research for logistics, scheduling and resource allocation

Geospatial Engineering

Satellite imagery, weather data and spatial datasets, turned into production intelligence.

Satellite data and remote sensing: Multispectral and SAR imagery, atmospheric correction to change detection
Weather and climate data: NWP outputs, reanalysis and climate risk modelling for energy, agriculture and insurance
Spatial analytics: Terrain modelling and location-based ML for logistics, real estate, telecom and urban planning
Earth observation platforms: Cloud-native GIS with STAC catalogues and analysis-ready data

Blockchain & Web3

Decentralised systems built to institutional and regulatory standards.

Smart contracts and protocol engineering: Formal verification, gas optimisation and security auditing
Asset tokenisation: Real-world assets with programmable compliance and fractional ownership
Enterprise blockchain: Permissioned networks for supply chain provenance and multi-party data sharing
DeFi and Web3 platforms: Institutional DeFi, dApps, wallet integration and governance
Digital identity and credentials: Self-sovereign identity, verifiable credentials and zero-knowledge proofs

Our own technology

Built here, and deployed.

Four programmes that began inside Scriptics with no customer attached, and are now systems in production.

01 Sequencing

AMR responsible-protein detection

Identifies the protein responsible for antimicrobial resistance from DNA sequencing samples, including where that protein is not in the reference catalogue.

Early cancer risk prediction

Cervical and breast cancer risk, identified up to three years ahead of clinical presentation. A risk model, designed to change screening interval and follow-up priority.

Radiology AI

Cancer detection from diagnostic imaging, at 96.4% accuracy against radiologist-adjudicated ground truth.

Aligner treatment planning

A manufacturable clear-aligner plan in a single day, on general-purpose CPU hardware, with the clinician inside the loop.

What connects them

Three habits.

Six practice areas and four owned technologies look like different businesses. They were not built that way.

Decide the compute budget first
Fix what the system is allowed to cost to run, then build inside it. We use GPUs where they earn their place; we do not let a design depend on them. It is why aligner planning runs on general-purpose CPUs, and why the diagnostic systems can go into a hospital basement rather than only into a cloud region.
Design the instrument and the software together
Sensor choice, data path and failure behaviour are model design decisions. Treat them as procurement and you end up with a model trained on data your shipped instrument does not actually produce.
Prove it in the workflow, not on a benchmark
A model that scores well on held-out data and then fails in the estate it was deployed into has been measured, not validated. We test against the workflow, the hardware and the case mix the customer actually has, because that is the only environment the result has to survive.

Start with the problem, not the product.

Tell us what has to be true for the system to work. We will tell you what we would build, and what we would leave alone.