Frontier AI · Eight models · One discipline

The world's best AI,
put to work.

Broadbase AI runs the frontier — routing every problem to the model that solves it best, cross-checking each answer against the rest, and engineering the prompts that turn raw intelligence into decisions. Working problems, bringing solutions.

Deep reasoning

GPT

Structured analysis and long-form synthesis across large bodies of research literature.

Explore the technology
The Technology

Eight frontier models. One discipline.

No single model is right for every problem. Each task is routed to the model that does it best — then cross-checked against the others so no single vendor's blind spot becomes ours.

Deep reasoning
GPT

Structured analysis and long-form synthesis across large bodies of research literature.

Careful analysis
Claude

Nuanced reading of policy and legal texts, where precision and caution matter most.

Multimodal
Gemini

Working across documents, imagery and data together — reports, maps, charts, CCTV stills.

In the workflow
Copilot

AI embedded directly in the everyday tools organisations already run on.

Efficient scale
DeepSeek

High-volume research triage at low cost — reading thousands of papers so we read the right ten.

Real-time signal
Grok

Live public-sentiment and fast-moving situational awareness during unfolding events.

Multilingual
Qwen

Reaching international front-line research that never gets translated into English.

Open weights
Llama

On-premise deployment where data cannot leave the building — sovereignty by design.

Prompt engineering is the craft that binds the stack. An ideal AI prompt is clear, specific, and contextual — it provides precise instructions while defining the desired output format. That discipline is the difference between an AI that generates text and an AI that generates decisions.

Cross-model validation is the safety net. Every significant finding is put to at least two independent models before it reaches a recommendation. Where they disagree, a human investigates. Where they agree, we still check the sources.

AI in Action

Eight working problems.

Each programme starts the same way: AI delving into publicly available research papers and the experience of national and international front-line organisations — and ends with an intervention a real organisation can afford to run. Each one links to a worked example.

CAP-01

Prompt Engineering

An ideal AI prompt is clear, specific, and contextual. We design and test the instruction sets that make frontier models produce decision-grade output — defined formats, cited sources, stated confidence.

LLM orchestrationOutput schemasEval loops
Explore the discipline
CAP-02

Anti-Social Behaviour

AI delves into publicly available research papers and national and international front-line organisations to develop targeted interventions with limited resources — evidence first, spend second.

Research miningMulti-model synthesisIntervention design
View worked example
CAP-03

Alcohol & Drug Management

A different approach to alcohol and drug-related anti-social behaviour — transforming how the Police manage the low-level crimes that fill police cells, freeing officers for the work only they can do.

Case-pattern analysisDiversion modelling
View worked example
CAP-04

Mobile Face Recognition & ID Fraud

Identity fraud costs the UK £1.8 billion annually. Mobile face-recognition puts real-time authentication in the field — matching a live face to a verified identity in seconds, on a device that fits in a pocket.

Computer visionEdge inferenceLiveness detection
View worked example
CAP-05

Civil Defence Planning

AI mines public research, real incidents at home and abroad, and the UK's existing resilience framework to design units of trained, insured, low-budget volunteers ready for major emergencies.

Scenario modellingResilience mapping
Read the blueprint
CAP-06

Harlow: Emergency Preparedness

Civil defence principles applied locally. The emergency services deal with the emergency itself; the civil defence team deals with the situations a large-scale event creates around it — a manpower injection when it matters.

Local deploymentVolunteer coordination
See the plan
CAP-07

AI Ethics

Keeping humanity safe, applying ethics to AI. Project Icarus is our alignment framework — twelve tenets that give the machine integrity before it is given power, and a mathematical, validate-then-attest method for proving every AI action was authorised by an accountable human.

Alignment frameworkValidate-then-attestDecision ledger
Read Project Icarus
CAP-08

Real-Time Recipe Cooking

Real time recipe cooking in your kitchen. Cookpit is our open JSON-LD cooking-file standard — validated, cryptographically signed recipes that coordinate starters, mains and desserts live, for the cook, chef, hobbyist or baker.

Open standardJSON-LD schemaEd25519 attestation
Start cooking
Innovation Riptide book cover — New techniques to energize and realize business potential, by Martin Jackson
The Book

Innovation Riptide

Martin Jackson's exploration of innovation methodologies challenges conventional thinking about creative processes and organisational transformation — an alternative foundation before AI came on scene, somewhere to launch your AI programme from.

12chapters
145pages
73graphics
Buy a copy
Contact

Get in touch.

Bring us a working problem. We'll bring back what the world's research — read by the world's best models — says about solving it.

Max 500 words