Chris J Mears

Projects

• LLM Evaluation Pipeline

  Systematically comparing distilled and quantized local models (Llama CPP, LM Studio) against API baselines — discovering that quantization introduces behavioral instability (repetition loops, hallucinations) beyond mere accuracy degradation, with implications for deploying LLMs in resource-constrained clinical NLP and medical document processing environments.

• Chicago Housing Predictive Visualization

  Full-stack analytics tool (D3.js, Leaflet, Svelte) forecasting housing flip probabilities from the Chicago public datasets.

• Open Brewery DB

  Creator and maintainer of an open-source dataset and API serving 800k+ weekly requests across 10+ years — the go-to resource for global brewery data with 70+ community contributors. Migrated from Rails/PostgreSQL to Laravel/SQLite with Meilisearch for real-time search, reducing infrastructure cost while scaling. Survived 10x traffic spikes through caching-first architecture. Currently building data quality pipelines to track brewery lifecycle events (openings, closures, relocations) across historical records.

• Brewery Review Analysis with Natural Language Processing (NLP) and Named Entity Recognition (NER)

  Springboard capstone — transformer-based NLP and NER pipeline for sentiment classification and entity extraction from brewery reviews; NER methodology directly transferable to clinical entity extraction.

• Credit Card Fraud Analysis and Prediction

  Built a machine learning pipeline (scikit-learn) to detect fraudulent transactions with severe class imbalance — optimized for recall over precision based on cost asymmetry (missing fraud is more expensive than investigating false positives), using SMOTE to improve minority class detection.