Kiryl Katushkin, FCCA
LinkedIn

Founded by an auditor who can build – and a data scientist who understands assurance

Most audit-analytics advice comes from one of two camps: auditors who can't build, or technologists who don't understand assurance. K Real Solutions was founded by Kiryl to sit in the overlap – FCCA-qualified, with an MSc in Data Science and fifteen years across financial services and Big 4 audit (EY, KPMG, Deloitte).

That experience includes putting AI and automation into live production inside a large, regulated financial-services firm – solutions that survive contact with real controls, real regulators and real audit committees. That's the difference between a proof-of-concept and something your function can actually run.

K Real Solutions is based in Edinburgh, UK, and works with clients across the UK and internationally, subject to local regulatory requirements

How we work

Built for regulated environments

Audit functions can't run on prototypes. Everything we deliver is designed to hold up under scrutiny – from your second line to your regulator.

Assurance-first

Designed by someone who has sat on both sides of the control – the analytics serve the assurance objective, not the other way round

Human in the loop

AI accelerates and widens coverage; auditors keep judgement and accountability. Explainable by design.

Works with your stack

Microsoft and Azure, Alteryx, Databricks, Snowflake – we build on what you already have and can govern, not a black box

Transfer, not lock-in

Your team owns the solution and the know-how. Documentation and upskilling are part of the deliverable.

Selected work

Assurance problems, solved in production

Drawn from delivery inside a large regulated financial-services firm. No client, system, dataset or document is ever named, reused or repurposed. Case studies are published in de-identified form only; third-party platforms are named, internal systems are not.

Continuous monitoring

Continuous controls monitoring platform

24/7Always-on monitoring
Challenge
Controls assurance relied on periodic, sample-based testing – long gaps and no live view of control health
Built
A monitoring platform in Microsoft Fabric: controls-performance feeds from multiple systems, blended via dataflows, with exception logic and combined scorecards in Power BI Service and Power Automate alerts the moment a control breaches
Microsoft FabricPower BIPower Automate
Manual control, automated

Unstructured email into a tested control

100%Population, not a sample
Challenge
Manual controls that hinge on reading inbound email get tested by sampling a handful – most of the population goes unchecked
Built
AI that turns an unstructured email feed into a structured dataset the moment mail lands, then reconciles it against the business record so the whole population is tested automatically. First applied to corporate-action notifications; the pattern fits any email-driven control.
Power AutomateAlteryxGenerative AIPower BI
Combined assurance

Firm-wide combined assurance map

4→1Assurance lines, one map
Challenge
Audit, risk, compliance and financial-crime teams assured risk in isolation – duplicated effort, blind spots at the seams
Built
A single top-down assurance map in Power BI showing coverage and work performed across every line, so each can target genuine gaps rather than re-cover ground
Power BISharePoint
GenAI · at scale

Bulk legal & document analysis

~3,000Documents analysed
Challenge
Large volumes of legal and client documents hold risk and insight no team can review manually at scale
Built
A reusable GenAI engine for bulk extraction and analysis. In one application it analysed ~3,000 client documents for ESG consistency, using cohort analysis to flag outliers where messaging over- or understated ESG considerations.
DatabricksGenerative AIRAG
GenAI · text

Policy contradiction review

WholePolicy set, read at once
Challenge
Large policy estates accumulate internal contradictions manual review rarely catches in full
Built
An LLM review that reads entire policy sets and flags inconsistencies and contradictions across long documents no reviewer could hold in their head at once
Generative AIText analysis
Vision AI · fraud

Fake-receipt detection with vision AI

EveryReceipt image inspected
Challenge
The “receipt required” control is easy to game – people attach blank pages that say “no receipt” just to clear the check
Built
A vision-AI indicator that inspects every uploaded image, separates genuine receipts from fabricated ones, and surfaces the highest-value cases, repeat claimants and the approvers signing them off
Vision AIPower BI
Process mining

End-to-end process mining

12-moEvent log reconstructed
Challenge
A creation, approval and distribution process was assumed to run one way; no one had seen how it actually ran
Built
Process mining across a 12-month event log – reconstructing the real process, its variants and rework loops, and quantifying the inefficiency
Process MiningEvent-log analysis
Automation

Automated audit-action follow-up

0Manual chasing
Challenge
Chasing findings and actions across stakeholders was manual and time-consuming, dragging on completion rates
Built
A Power Automate solution that notifies auditors and stakeholders of upcoming and overdue actions with full context, and opens the follow-up conversation automatically
Power AutomateTeams
GenAI · briefings

Automated intelligence briefings

AnySource → scheduled digest
Challenge
Insight that would sharpen audit – incidents, control breaches, emerging risks, market news – sits scattered across systems and reaches the team late, or not at all
Built
Briefings that pull from any source, have AI summarise and categorise, and land as a structured daily or weekly digest. Incident summaries were the first; the same engine drives any feed the team needs to stay ahead of.
Power AutomateAlteryxGenerative AI
Further workPDF authorisation & SoD testing · Alteryx + PythonControl-test generator · LLMAudit QA challenge & sentiment · LLMAutomated risk assessment · LLM
StackMicrosoft FabricPower BIPower AutomateAlteryxOpenAIAnthropicGeminiAzure DatabricksSnowflakePythonRAG pipelines