Senior Data Scientist

Sameer
Maurya

Cybersecurity · AI Governance · Data Science

Almost a decade building AI systems that regulators can audit and executives can trust, focused on financial compliance, model risk, and responsible AI.

Start Here About me →
Sameer Maurya
Scroll to explore
[01] Selected Work All projects →

Work that holds up under review.

LLM Eval
Model Risk · LLMs

LLM Eval Framework

I built a validation suite that turns raw LLM outputs into evidence regulators can audit.

Live on HuggingFace ↗

LLM Eval Framework

A local-first auditing suite that automates Model Risk Management artifacts regulators require — converting raw LLM outputs into compliant evidence.

  • Adversarial Red-Teaming
  • SHAP / LIME Attribution
  • SR 11-7 Mapping
  • EU AI Act Ready
Live on HuggingFace
meerax
OSS Package · Python

meerax

I packaged my ML tooling into one installable toolkit that every project here is built on.

meerax

One installable Python package every project in this portfolio is built on, with a shared governance checklist every one of them passes before release.

  • LLM Provider Abstractions
  • Eval + Report Builder
  • Scaffolding CLI
  • 158+ Tests, mypy Strict
View on GitHub
Pixel Drift
Computer Vision · CycleGAN

Pixel Drift

I extended a 3-channel CycleGAN to any two image domains with a channel-tiling wrapper.

Pixel Drift

Replaces tensorflow_examples' hardcoded 3-channel assumption with a channel-tiling wrapper, so the same architecture extends to any two-domain image task.

  • TensorFlow CycleGAN
  • Channel-Tiling Wrapper
  • Any Two Domains
  • Apple↔Orange Reference Pair
View on GitHub
9+ Years building production ML
90% Less manual review effort with LLM summarisation
3 Regulatory frameworks mapped: SR 11-7, EU AI Act, GDPR
12 Projects built, from GenAI to classic ML
[02] Problems I Solve

The problems I'm built for.

[01]

Validating LLMs for regulated finance

Banks need evidence, not demos. I build validation suites that turn raw model outputs into artifacts that model-risk reviewers and regulators can audit.

SR 11-7 EU AI Act Red-teaming SHAP / LIME
See the project → [02]

Making model decisions explainable

A flagged transaction or a recommendation is only useful if someone can say why. I pair models with plain-English explanations and a documented audit trail.

Fraud detection GenAI narration Governance Fairness
See the project → [03]

Shipping ML that stays shippable

Prototypes die in notebooks. I turn them into packaged, tested projects on one shared toolkit, with a release checklist every project has to pass.

meerax CI / CD Docker Typed Python
See the project →

Building AI that has to stand up to regulators? Let's talk.

Say hello →