Natanel Richey

• Built and evaluated a multi-class infant-cry classification pipeline, focusing on robust validation, interpretability, and reproducibility.
• Developed end-to-end ML evaluation workflows in Python
• Ran rigorous, comparative model testing using nested CV with fold-level and aggregated analysis (macro/micro F1, per-class precision/recall/F1, confusion matrices, latency) to reliably determine which approaches generalized best across splits
• Built the full training pipeline combining audio preprocessing (spectrogram-based features / AST embeddings), data augmentation, feature construction (late-fusion of embedding + acoustic features), and model training/inference across multiple model families (e.g., MLP, LightGBM, plus classical baselines like SVM/Nearest Centroid where applicable)
• Engineered prompt variants for LLM classification (Chain-of-Thought, Tree-of-Thought variants), systematically compared API responses across providers, evaluating reasoning quality, and refining prompt structure and routing logic to reduce token costs while refining performance

• Founded and lead a company specializing in curing operational chaos for small businesses through personalized technical solutions.
• Design and implement website automation, integration, and workflow optimization tools that eliminate wasted effort and replace scattered workflows with unified operating systems.
• Bridge the gap between business needs and cutting-edge technology, specializing in Web/App Development and Applied Machine Learning.
• Create tailored technical solutions that fit naturally into client workflows, focusing on efficiency, automation, and organization.
• Work directly with clients to understand their unique business processes and deliver solutions that bring measurable success.

Languages:

Skills: