AI & risk advisory for credit unions and community banks

Practical AI. Examiner-ready from day one.

We help credit unions, community banks and other smaller lenders put AI to work in credit decisioning, line management, reporting, AML/BSA and back-office operations, and we govern every tool like any other model: measured, documented, monitored and owned by your people.

Clarity in every dimension.

See what we do ↗
DATA & QUESTIONSDECISIONS & EVIDENCE
FOLD THE PATH. KEEP THE PICTURE UPRIGHT.
BUILT AROUND YOUR INSTITUTIONMEASURED. DOCUMENTED. OWNED BY YOUR TEAM.
What we do

Six places AI pays off.
Each one governed.

We combine domain expertise with large language model (LLM) tools and conventional statistical methods, using whichever fits the problem better. LLMs draft, extract, summarize and recommend. Credit, compliance and customer decisions stay with accountable people at your institution.

01 / CREDIT DECISIONING

Faster, more consistent credit decisions

Scorecards, decision rules and credit policy for consumer, small-business and commercial lending, with challenger testing.

Where AI helps
  • Extracting data from financial statements, tax returns and bank statements
  • Drafting credit memos and committee summaries
02 / LINE MANAGEMENT

Line assignment and limit strategy

Initial line, line increase and line decrease strategies, risk/reward optimization and controlled experiments.

Where AI helps
  • Rationale text for adverse action and review files
  • Monitoring narratives
03 / REPORTING & ANALYTICS

Board and committee reporting

Board, ALCO and credit committee packs, CECL allowance support, portfolio and concentration reporting, and stress testing.

Where AI helps
  • First-draft commentary from query results
  • Natural-language access to approved data
04 / AML/BSA & FRAUD

Sharper monitoring, faster casework

Transaction-monitoring tuning, alert triage, KYC/CDD refresh and model-based detection.

Where AI helps
  • Alert summaries and SAR narrative drafts for analyst review
  • Adverse-media screening
05 / OPERATIONS

Back-office workflow automation

Mapping and redesign of loan servicing, account opening, exception handling and back-office workflows.

Where AI helps
  • Document intake and classification
  • Procedure Q&A assistants, ticket and email routing
06 / MODEL RISK & AI GOVERNANCE

Governance for every model you run

Model inventory, validation, ongoing performance monitoring, AI use policy and vendor model reviews.

Where AI helps
  • Documentation drafting and evidence collection
  • Control-testing support

For credit unions

Line management, member-service volume, NCUA reporting and collections prioritization, with automation and governance sized to your team and your member mission.

For community banks and other lenders

Credit memos and spreading, consistent underwriting, CECL and stress testing, and third-party oversight for institutions, CDFIs and specialty lenders without a dedicated data science team.

Why Risk Prism AI

Fold the path.
Keep the picture upright.

Our mark is a Porro prism, the optic inside a pair of binoculars. It adds nothing to the light. It folds a long optical path into a compact instrument and turns an inverted image right side up, so what you see is sharp, magnified and the right way round.

That is the job we take on. AI shortens the path: language models and statistical methods applied only where they measurably save time or improve decisions. Risk discipline keeps the picture upright: every tool is treated as a model under your risk framework, tested against ground truth, documented for examiners and owned by your staff.

01

Assess 2–4 weeks

Map processes, data, systems and controls. Find where automation pays off and where it adds risk.

Deliverable: a prioritized opportunity map with estimated time savings and risk impact.

02

Pilot 6–10 weeks

Build one solution on a contained use case with real data, run it in parallel with the current process, and measure accuracy, cycle time and exception rates.

Deliverable: a working pilot, validation results and governance documentation.

03

Deploy

Integrate with your core, loan origination, AML and reporting platforms. Train users; write procedures and controls.

Deliverable: a production solution and operating runbook.

04

Monitor and transfer

Ongoing performance monitoring, periodic recalibration and annual review support, with knowledge transfer so your staff own the solution.

You make a go/no-go decision at the end of every phase.

AI you can defend

Built to be explained
to examiners and boards.

Every LLM-based or statistical tool we deliver is treated as a model or critical process under your risk management framework, scaled to your size and to the tool's materiality. Documentation, audit trails and monitoring are built in from the start, not retrofitted before an exam.

MEASURED

Tested against
ground truth

LLM output is scored with quantitative metrics (accuracy, error rates, calibrated uncertainty intervals) before and after deployment. It is not judged by impression, and not by another LLM alone.

ACCOUNTABLE

People decide.
AI assists.

No credit denial rests on an LLM alone, and analysts keep the call on alert dispositions and SAR filings. Credit solutions are tested for disparate impact and support specific, accurate adverse action reasons under ECOA/Regulation B and FCRA.

CONTROLLED

Documented, monitored,
protected

Plain-language documentation of what each tool does, its limitations and its controls. Customer data is handled under GLBA and your information security program, and it is never used to train third-party models without your written consent.

Vendor-neutral, data stays home

We work with the platforms you already run and favor deployments that keep customer data in your environment or an approved private cloud, using commercial or open-weight models as data-protection and cost requirements dictate. We accept no undisclosed referral fees or commissions from vendors.

Scoped to your budget, built to hand over

Fixed-fee assessments and pilots, project fees for deployment, and optional retainers for monitoring, validation or fractional chief risk and analytics officer support. The principals assigned to your engagement do the work, then train your staff to run it without us.

Our principals

Senior risk experience.
Hands-on engineering.

Between us, more than 60 years of senior risk, analytics, AML and transformation experience at global systemically important banks, together with hands-on software engineering. We know what examiners expect of a model or a process, and we build the working tool ourselves.

DM

Dmitriy Mayorov

Credit risk modeling & AI evaluation

Dmitriy has spent more than 20 years in credit risk: PD, LGD, EAD and rating-transition models feeding CCAR, CECL, IFRS 9 and Basel capital. Most recently he led a distributed team owning 21 wholesale credit models at a global bank and answered Federal Reserve and OCC inquiries on modeling decisions. Earlier he designed card credit-line policy, including line decrease strategies. An engineer and software developer before he moved into risk, he builds the tools himself, including perEval, an open-source suite that objectively tests whether AI agents can build and validate credit loss models.

LT

Laura Thomae

AI risk & transformation

Laura has worked in AI since the expert-system era, starting as an AI engineer at Harris Corporation and in the AI Lab at S.W.I.F.T. She is Head of AI Risk and Compliance at Freshriver.ai, and was previously a Managing Director in Risk Management at two global banks, leading regulatory remediation, data and reporting transformation, and technology rationalization in organizations of 1,800+ staff. She makes sure what we build fits your capacity, gets adopted and becomes part of day-to-day operations.

RR

Raji Ramachandran

Model analytics, AML & AI operations

Raji is a hands-on quantitative developer and software engineer with more than 25 years in financial services. At a global bank she led an 18-person team building machine-learning AML detection models, led continuous monitoring and review of 150+ forecasting, risk-capital and stress-testing models, closed two regulatory MRA remediations, and built an automated model-monitoring toolkit. A Ph.D. economist, M.S. computer scientist and co-inventor on a granted data-quality patent, she turns policies into the controls, monitoring and workflows that make AI auditable.

About these profiles

Profiles summarize each principal's professional background from their résumés and public biography. The roles described were held at former employers and are not Risk Prism AI client engagements.

Laura: professional backgroundDmitriy: analytical work
Research & white papers

Ideas you can examine.
Methods you can run.

Papers, code and practical perspectives on measuring and governing AI in financial risk. Explore the analytical methods and engineering behind our work.

Research paper & R package

Credit loss modeling with macroeconomic factors

Research extending the Vasicek credit loss model, accompanied by the vasicekfit R package for exploring the method.

Open-source evaluation code

Objective evaluation of analytical AI agents

perEval tests whether AI agents can develop, estimate and validate credit loss models, scored against a known reference with prediction intervals rather than by LLM judges.

GitHub portfolio · research in development

Governed workflows & analytical prototypes

Raji's portfolio covers agent controls, resumable workflows, data pipelines and expected-loss research. Project scope and maturity vary.

These are the principals' individual resources, not Risk Prism AI white papers. Read research and prototypes in the context of their documentation, limitations and licenses.

Proposed white paper topics

Developing our perspective

Topics for future Risk Prism AI papers; publications and companion code will be linked here as they become available.

  • Measuring LLM output against ground truth
  • Governing LLM tools under model risk management
  • Right-sizing AI governance for community institutions
  • Fair lending and adverse action when AI drafts the file
Start a conversation

What would you
automate first?

Bring us one process that takes too long, or one model you cannot fully explain to your examiner. We will scope a fixed-fee assessment or pilot around it, building on the systems you have today.