Platform
An AI-powered risk intelligence system that captures the Actual Process—not just the Official Record. Built to amplify human capability, not replace it.
A risk analyst might review 50-100 scenarios per quarter—not because they're lazy, but because that's all the system allows. AI changes the equation: 10X or even 100X more than we can achieve today, with humans making the final decisions.
Agentic systems don't replace your core systems—they augment them, filling the gaps and expanding what's possible.
Your existing infrastructure: market risk engines, credit workflow systems, regulatory reporting tools. Deterministic, auditable, embedded in the organisation.
AI agents that run tools in a loop to achieve goals. They operate alongside core systems, capturing what those systems miss.
The key insight: Agentic systems make core systems more complete. They capture the discussions, research, and reasoning that happen around the formal process—and store it in context graphs that make AI actually useful.
A simplified three-layer architecture that finally captures the Actual Process
Your existing infrastructure stays in place. Credit workflow tools, market risk engines, regulatory reporting, trade capture systems. These continue to provide the Official Record—deterministic, auditable, unchanged. They're good at what they do.
Risk Agents provides: Integration patterns and examples. You build: Connections to your specific systems.
This is the new layer that captures the Actual Process. It comprises two parts:
The Foundation
The core agentic loop provided by Large Language Models—a general-purpose reasoning engine trained on the internet. On its own, nothing more than that.
You purchase: LLM subscriptions from major labs (Anthropic, OpenAI, etc.)
Skills & Patterns
Composable building blocks of prompts and supporting code that provide specific business context. This is what transforms a generic reasoning engine into a specialist risk management tool.
Risk Agents provides: Skills library and patterns. You customise: Organisation-specific adaptations.
Critically, business owns these skills and updates them directly—changes happen in days, not months.
This serves both Core Systems and Agentic Systems:
Structured Data
Business facts with key attributes—trades, counterparties, transactions. Feeds both layers.
Unstructured Data
Meeting transcripts, Slack/Teams discussions, decision evidence, email threads. This is where the Actual Process lives.
Context Graphs
Accumulated reasoning and decision traces. Captures not just what decisions were made, but why, with what reasoning, under what circumstances.
You build: Your own data connections, context graphs from your decision traces. This is your competitive advantage.
Practical examples showing how the platform works for real risk management tasks
End-to-end credit approval using composable skills: Credit Request, Legal Review, Credit Analysis (with sub-skills for research, financial spreading, climate scorecard).
Skills built: Climate Scorecard Filler ✓
Skills in roadmap: Credit Request, Legal Review, Financial Spreading
AI-assisted scenario development following the annual stress testing cycle. Covers pillar stress and point-of-weakness scenarios across all asset classes.
Skills built: Stress Scenario Suggester ✓, Pillar Stress Generator ✓
Continuous monitoring of regulatory changes with automatic impact assessment on the Risk Taxonomy. Updates skills when requirements change.
Built: Regulatory Risk Researcher agent ✓, Regulatory Change Assessor skill ✓
Structured AI-guided goal review process. Replaces ad-hoc "offsite slide deck" approach with systematic assessment of status, gaps, and initiatives.
Planned: Horizons Reviewer skill (adapted for business), Weekly Review skill for monitoring
AI excels at documentation—it doesn't tire of it. Specialist skills maintain consistency across risk disciplines and keep documentation current.
Planned: Policy-updater, procedure-writer, and layer-specific documentation skills
General Q&A across your risk documentation, guided by the taxonomy. Progressive disclosure means agents only load relevant context—efficient and accurate.
How it works: Taxonomy acts as a "map" for where to search, avoiding the need to load all documentation into memory.
Move from standardised templates to dynamic, natural language reports generated from underlying data. Ask questions, get answers with full context.
Learn more about reporting →Why decision traces matter more than data
"The taxonomy is the map; the context graph captures how people actually navigate that map."
Every system has a state clock—what's true right now—and an event clock—what happened, in what order, with what reasoning.
We've built trillion-dollar infrastructure for the state clock. The event clock barely exists. That's the gap context graphs fill.
Over time, context graphs become world models. The organisation can simulate "what if?"—not just retrieve "what was."
That's the difference between a chatbot and a trusted risk advisor. Simulation is the test of understanding.
The model is the engine. The context graph is the world model that makes the engine useful.
Risk Agents is a community, not just software. Here's the split.
Context provides the foundation. Skills provide capabilities. Patterns orchestrate workflows. Goal Alignment ensures everything connects to strategic outcomes.
Join as a founding member to help shape the platform and get early access.