Keryx/Solutions
Outcomes — not packages

Where the decisions your business runs on actually live.

Solutions are organised around the outcome you need delivered, not the size of the engagement. Every page maps to architecture, delivery approach and reference engagements — never bronze, silver or gold.

01 — Enterprise AI architecture

Architecture for AI that has to defend itself.

The reference architecture, target operating model and 12-month roadmap for your enterprise AI estate — sized to survive your own risk function.

Challenge

You have AI experiments. You don't yet have an architecture that the second line, your auditors and your board can all live with.

Strategic perspective

The architecture is the artefact. Models change. Vendors change. The architecture is what gets governed, defended and reused.

Architecture view

Native execution. Tenant-isolated decisioning. Policy-as-code governance. Audit ledger as a first-class system, not a logging side-effect.

Delivery approach

Six weeks, founder-led, three artefacts: target architecture, operating model, sequenced 12-month roadmap.

Reference examples

  • Tier-1 retail bank · UK · enterprise AI architecture
  • Global asset manager · cross-region target operating model
02 — AI governance & control

Governance that survives an audit.

Policy-as-code, approval flows and evidence packs that map to SR 11-7, SS1/23, Consumer Duty, the EU AI Act and DORA — without bolting paperwork onto an unbuilt foundation.

Challenge

Your board has approved AI in principle. Your second line has approved it in PowerPoint. Neither has approved it in production.

Strategic perspective

Governance is architecture, not a committee. Lineage, attestations and approvals belong inside the system, not in a quarterly review.

Architecture view

Policy DSL, attestations as code, evidence packs generated per decision, regulator-API exposure for case-by-case interrogation.

Delivery approach

Twelve weeks alongside your second line. Joint authorship of the policy library. Dry-run audits before any model goes live.

03 — Intelligent automation

Decisions in motion. Audit trails to match.

Where a decision triggers an action — credit approval, claims triage, KYC classification, suitability — we wrap the action in the same governance as the decision.

Challenge

The robotic-automation estate is brittle. The new agentic estate is unaccountable. Neither is operationally safe.

Strategic perspective

Automation is a special case of decisioning. It needs the same drift, the same explainability, the same audit ledger.

Architecture view

Decision Mesh + agent execution layer + HITL queue + step-level approval. One evidence chain, end to end.

Delivery approach

One workflow at a time. We instrument the existing workflow first, then automate, then govern.

04 — Modern data platforms

The platform underneath the decisions.

Lakehouse, Iceberg, native query, lineage, contracts. The data layer that operational AI requires to be auditable, performant and tenant-safe.

Challenge

Your data estate was built for analytics. You're now asking it to support decisions in milliseconds, with regulator-grade lineage.

Strategic perspective

Lakehouse is not a destination, it's a substrate. The interesting decisions are about contracts, tenancy and the operating model.

Architecture view

Iceberg-native lakehouse, contract-versioned data products, lineage across the operational/analytical boundary.

Delivery approach

One data product, one decision, one tenant — then scale the pattern.

05 — AI operating models

Who owns what, when AI runs in production.

Roles, RACI, second-line/first-line interfaces, model lifecycle ownership and incident response — designed for an organisation where AI now decides, not just predicts.

Challenge

Centre of excellence vs federated. Risk owns it vs business owns it. The org chart hasn't yet caught up with the architecture.

Strategic perspective

The operating model is downstream of the architecture. We design them together, in the same room, with the same artefact.

Architecture view

Federated execution, central governance, line-aligned product ownership. The mesh enforces the model in code.

Delivery approach

Workshops with line, risk, audit, security and CIO together. Two weeks. One operating-model artefact. Signed by all five.

06 — Agentic workflows

Agents that an audit committee will sign.

Tool-calling, MCP, planner-executor patterns, step-level approval, full evidence chain — engineered for production rather than for keynotes.

Challenge

Agentic systems make decisions you can't replay. The next regulatory question will be: who approved that?

Strategic perspective

An agent is a chain of decisions. Each one needs the governance you'd require of a human at the same desk.

Architecture view

MCP-bridged tools, step-level approvals, replayable agent traces, drift on planner outputs and on tool outcomes.

Delivery approach

Agentic workflows are productised inside Keryx Labs first, then applied to the engagement. We don't ship demos.

07 — Data product strategy

From dashboards to data products.

Data products with contracts, owners, SLAs and operational consumers — not BI deliverables with a refreshed name.

Challenge

You have a data mesh slide. You don't yet have a data product anyone has staked their bonus on.

Strategic perspective

A data product is the unit of accountability. It has an owner, a contract, an SLA and a consumer who can break it.

Architecture view

Contract-versioned data products, native lineage, operational and analytical consumers, mesh-attached governance.

Delivery approach

Three data products, end-to-end, in eight weeks. Pattern, then scale.

Strategic implementation partnerships

Bring us an outcome. We'll bring the architecture.

Tell us the decision your business depends on, and which of these outcomes is in scope. We'll come back with a 90-day shape — your cloud, your controls, your risk function in the room.

Talk to us → See it in production