Detection is half the job.
Brontë detects material risks across the financial services ecosystem. This prototype shows what happens next — an AI agent that takes a detected control gap and ships the remediation work end-to-end, with human-in-the-loop approval and a full audit trail.
Workflow
Diagnose, plan, draft, validate, and evidence the fix in one controlled flow.
Approval model
No execution without a named control owner and explicit approval checkpoint.
Output
Generates operational artefacts a real bank team could inspect, refine, and file.
What the prototype demonstrates
A remediation agent designed like an engineering system.
The point is not that AI can write text. The point is that detected gaps can be translated into bounded work, explicit approvals, and audit-ready evidence with far less manual coordination.
Upstream input
Detected control gap
The agent starts with a structured export from a detection platform, not an open-ended chat prompt.
Decision model
Bounded remediation planning
Every proposed fix includes blast radius, reversibility, and a named approval gate before anything moves.
Final output
Audit-ready artefacts
The result is operational work product: Jira, Slack approval, validation logic, and regulator-friendly evidence.
Concept Overview
The gap this fills.
Detect.
Modern risk platforms are good at finding broken controls. They map risks to controls, run continuous tests, and surface gaps in real time. This part is increasingly solved.
Report.
Once a gap is found, most platforms route a ticket to a human and stop there. The gap is logged, dashboards update, leadership sees a number go up. The risk itself is unchanged.
Act and remove.
This is where the work actually lives. Diagnosing the root cause, designing a right-sized fix, drafting the change, getting approval, validating the result, and producing audit-ready evidence. Today this falls on tired humans. It doesn't have to.
Execution Model
How the agent works
Receive
Agent ingests a structured gap export from an upstream detection platform like Brontë.
Diagnose
Reads the control definition, the evidence, and the broader policy context.
Plan
Drafts a right-sized remediation plan with clear blast radius and human approval gates.
Ship
Generates the executable artefacts: Jira ticket, Slack approval, change spec, validation test.
Prove
Produces the post-fix evidence artefact for audit, board, and regulator review.
Guardrails
Built on real engineering principles.
Blast radius first.
Every action the agent takes is scoped, bounded, and reversible. Permissions are explicit, not inherited.
Human in the loop.
The agent never executes a change without an approval gate from a named control owner.
Auditable by default.
Every reasoning step, tool call, and decision is logged with full traceability for regulators and internal audit.
About The Builder
Built by an engineer who ships the hard part.
Bernard Adjei-Yeboah
AI engineer, product builder, and systems operator.
I'm a Sydney-based full-stack engineer and AI solutions architect. I built this prototype in response to Newton Russell's Forward Deployed Engineering brief to show how I think about regulated workflows, bounded automation, and real-world remediation systems.
Through EchoFlow Labs, I use a demo-build-sell approach to ship production-grade AI products across workflow automation, copilots, analytics, and operational tooling. In my current work at HEALTHiClinic, I lead technology delivery across clinical NLP and multi-system coordination platforms.
My background combines engineering delivery, applied AI, and systems assurance: a Master of Information Technology in Artificial Intelligence, nine ACS professional accreditations, and hands-on experience shipping products that need to work in production rather than just in slides.
5+
Years shipping AI systems
13+
AI products shipped
9
ACS accreditations
MIT
Artificial Intelligence
Full-stack AI delivery
I design the product, implement the application, wire the model layer, and ship the production system end-to-end.
Risk and workflow thinking
I focus on bounded automation, approval gates, evidence trails, and operator trust instead of novelty for its own sake.
Systems integration
My production work spans TypeScript, Python, React, Next.js, NLP pipelines, and multi-system operational tooling.
Demo-build-sell execution
I prefer showing working artefacts quickly, then iterating against reality instead of presenting abstract claims.
Disclaimer
Important context for this prototype.
This prototype is presented as a portfolio artefact and product demonstration. It is intended to make the underlying engineering judgment inspectable, not to imply live deployment, client usage, regulatory approval, or commercial endorsement.
Independent portfolio prototype
This site was independently designed and developed by Bernard Adjei-Yeboah as a working demonstration of product thinking, risk engineering, and AI-agent execution.
No implied affiliation
Unless explicitly stated, this prototype is not a Newton Russell, Brontë, or customer product and should not be interpreted as endorsed by, or representative of, any organisation referenced here.
Illustrative artefacts
Example control gaps, remediation plans, approvals, evidence packs, metrics, and workflow outputs may be simulated, synthesized, or simplified for demonstration purposes.
Not professional advice
Nothing on this site constitutes legal, regulatory, audit, security, investment, or other professional advice. Real production use requires independent review and domain approval.
Human review remains mandatory
Any remediation pattern shown here should be treated as a product concept. In a live environment, changes would require validated controls, access governance, testing, and accountable human sign-off.
Company names, product names, and market references remain the property of their respective owners and are used only to explain the operating context of the concept. Any real implementation of a remediation agent in a regulated environment would require institution-specific policies, legal review, security controls, model governance, and production change management.