Senior Financial Regulatory Professional · Independent SupTech Researcher

I work in financial supervision and build AI-assisted tools to solve problems regulators actually face, from evaluating institutional complaint responses to detecting anomalies in regulatory datasets.

My focus: practical SupTech that is explainable, governed, and deployable within real supervisory constraints.

• AML/CFT risk intelligence — typology analysis and risk-scoring systems for supervisory use
• Conduct risk & complaint analytics — NLP-based evaluation of institutional response quality
• RAG pipelines for regulatory knowledge — retrieval-augmented systems for policy and rulebook Q&A
• Supervisory data quality — anomaly detection and validation frameworks for regulatory filings
• AI governance in financial regulation — explainability, audit trails, and human-in-the-loop design

Problem: Supervisors spend significant time manually reviewing institutional complaint responses for evasiveness or poor quality — a process that does not scale across large complaint volumes.

What it does: Applies NLP classification to public CFPB consumer complaint data to automatically flag potentially low-quality or evasive institutional responses, enabling faster triage for conduct-risk analysis.

Scope: Tested on CFPB public dataset spanning multiple product categories and financial institutions.

Stack: Python · NLP / Transformers · Pandas · Regulatory Analytics

Problem: Financial datasets submitted to regulators often contain systematic errors and reporting gaps that go undetected until examination creating risk for both institutions and supervisors.

What it does: An AI-assisted audit platform that identifies anomalies, inconsistencies, and data quality issues in financial and regulatory datasets, designed with supervisory use cases in mind.

Design principle: Outputs are structured to be explainable and audit-trailed built for environments where a supervisor must be able to justify every flagged item.

Stack: Python · Statistical Anomaly Detection · Data Quality Frameworks · Risk Analytics

• Python — data pipelines, NLP preprocessing, model integration (primary language across all projects)
• NLP & LLMs — text classification, entity recognition, RAG pipeline design for regulatory and legal text
• Regulatory data — CFPB public datasets, synthetic supervisory data; familiar with financial reporting formats
• Anomaly detection — rule-based and statistical methods for structured financial dataset validation
• AI governance — explainability design, human-in-the-loop checkpoints, audit trail architecture for supervised ML

AI in supervision should augment, not replace, regulatory judgment.

That means building tools that are explainable to examiners, proportionate to institutional risk, and accountable when they get it wrong. The four principles I return to:

• Explainability — supervisors must be able to understand and defend every output
• Governance — AI tools need oversight structures, not just accuracy metrics
• Proportionality — the level of automation should match the level of supervisory risk
• Accountability — when a tool flags something incorrectly, there must be a clear path to correction

Supervisory capacity is the goal. AI is one path toward it.

• LinkedIn: linkedin.com/in/razabakirshah
• Open to: Research collaborations · SupTech and RegTech conversations · Feedback from practitioners building in this space

If you're working on AI tools for financial regulators — or on regulating AI in finance — I'm interested in the conversation.

Note: All projects use public or synthetic data only and do not represent the systems, methodologies, or positions of any employer, regulator, or institution. No confidential or restricted supervisory information is used in any repository.