Sameer Maurya
Data Scientist — Generative AI · LLMs · Model Risk Management
8+ years building AI systems at the intersection of deep learning, financial compliance, and production engineering — from LLM validation at Bank of America to real-time OCR pipelines and influencer recommendation engines.
I specialise in Generative AI and Large Language Model implementation, with a strong focus on responsible deployment in regulated environments. My work spans model development, risk governance (SR 11-7), cybersecurity, and the full ML lifecycle from research to production.
This site is where I publish technical deep-dives — structured analysis on AI regulation, LLM applications, Model Risk Management, and financial technology. Not hot takes — real breakdowns of what's changing and why it matters.
Core Expertise
- Generative AI LLM implementation, fine-tuning, and production deployment
- Model Risk SR 11-7 compliance, MRM diagnostics, validation frameworks
- Computer Vision Deep learning pipelines, OCR, facial recognition, edge deployment
- NLP NER, intent detection, summarisation, document intelligence
- Responsible AI Cybersecurity, bias detection, adversarial testing
- Cloud & Infra AWS · Azure · Heroku · Python · PyTorch · TensorFlow
Experience
- Developed and validated LLM solutions (BART, GPT, Llama) for financial services
- BART-based summarisation — reduced manual review effort by 90%
- Engineered Identity Threat Detection & Response (ITDR) models
- Adversarial robustness testing and vulnerability assessments for AI systems
- Built NER and intent detection features for conversational bot platform
- Developed OCR services for identity document recognition at scale
- Created unified ML API powering chatbot products organisation-wide
- Built recommendation system for influencer selection
- Developed age and gender prediction models using Keras
- Presented campaign analytics and model outcomes to executive stakeholders
- Led research team on image preprocessing pipeline for edge deployment
- Facial recognition attendance system — 97.8% accuracy
- Implemented dense image captioning system