Bernard Griffin

About

ML/NLP practitioner with experience building a transformer-failure prediction system that combines sensor signals with unstructured maintenance comments using LLM-based extraction and similarity validation. Strong emphasis on production readiness—data leakage controls, SQL-driven data quality tiers, and rigorous bias/fairness validation (including contract/spec evaluation across diverse company profiles).

Experience

Education

Key Strengths

• Built predictive model for transformer failure using sensor data plus maintenance comments
• Applied LLM-based extraction and sentence chunking to structure unstructured maintenance text
• Used similarity metrics (cosine) to validate text/embedding proximity
• Strong focus on preventing data leakage (placeholders during sensor downtime)
• Baseline modeling with feature importance to refine feature set
• Designed data quality tiers (gold/silver/bronze) via SQL for production modeling
• Emphasis on production reliability: guardrails, hallucination checks, bias/fairness validation, stress testing
• Experience validating model fairness across different company sizes and demographics in contract/spec evaluation