Charles Rice — Senior Data Scientist

Charles studied mathematics at Wayne State University and later did graduate coursework and research at Portland State University in neural networks, reinforcement learning, and genetic algorithms. His research included evolvable hardware and automated stock trading systems.

Before moving into Data Science, Charles spent over two decades in software engineering. He built iOS, tvOS, and watchOS applications, developed full-stack systems, and worked with companies like Ford Motor Company, Btrieve, Tektronix, Nike, Xerox, and Hewlett-Packard. He shipped ten apps to the App Store and contributed to products like an ODBC driver for Btrieve.

For the past five plus years, Charles has focused on Data Science and Machine Learning. He's built AI agent systems, fine-tuned LLMs, developed computer vision models for cryo-electron microscopy, and tackled NLP problems ranging from topic modeling to contact data cleanup. His work spans PyTorch, scikit-learn, HuggingFace, LangGraph, and vector databases—taking projects from prototypes through production deployment.

Charles grew up in Detroit where he was active in the punk and electronic music scenes. He played guitar and keyboards and built his own guitar pedals. Music remains central to his life—he's an active listener and supporter of WFMU, the legendary freeform radio station, and maintains a collection on Bandcamp.

With his background in mathematics, machine learning, and decades of programming experience, Charles continues to explore new possibilities in Data Science, AI, and Machine Learning.

Data Science

Charles has 5+ years of experience in data science and machine learning, supported by 25 years of software engineering. His work covers the full project lifecycle from data exploration and modeling through production deployment. He's built solutions using PyTorch, scikit-learn, TensorFlow, and the standard data science stack including Pandas, NumPy, and Matplotlib.

He's developed classification models for email engagement detection with Snowflake dashboards for production monitoring, analyzed agricultural datasets to assess market viability for new business initiatives, and built XGBoost models using 3D body scan measurements for clothing size recommendations. He's also designed PostgreSQL databases for metadata management and implemented MLflow as a standardized framework for model development and deployment.

Natural Language Processing (NLP)

Charles has worked on NLP projects spanning topic modeling, information extraction, and text classification. He designed a contact cleanup system that combined traditional NLP techniques, table lookup, and LLM fine-tuning to demonstrate feasibility for production deployment.

He conducted topic modeling on reviews of herbal remedies that reduced analysis time from days to hours. He's also applied NLP methods to extract structured information from unstructured text in large government datasets, making them more usable for downstream analytics.

His NLP work uses tools including spaCy, Sentence Transformers, and Gensim, focusing on practical applications that solve specific data problems.

Computer Vision

Charles solved a Cryo-EM object detection problem using transfer learning, which removed a bottleneck in product development. He then designed an automation system for Cryo-EM data acquisition and implemented K-nearest neighbors optimization for multi-image target acquisition, reducing acquisition time significantly.

He's also built image classifiers for various applications, including a kaggle competition where he placed 22nd in kitchenware classification. His computer vision work uses PyTorch, TensorFlow, OpenCV, and Scikit-Image, with Albumentations for data augmentation.

He designed and managed implementation of a semi-supervised learning system that reduced annotation time from days to hours.

Generative AI

Charles has worked with large language models through OpenAI API, HuggingFace, and Ollama. He fine-tuned LLMs using QLoRA and LoRA for domain-specific applications including email sequence generation. He's implemented RAG (Retrieval-Augmented Generation) systems using vector databases including pgvector, Chroma, and FAISS.

His generative AI work focuses on practical applications: fine-tuning models for specific tasks, implementing effective prompting strategies, and integrating LLMs into larger systems.

AI Agents

Charles has built multi-agent systems using LangGraph. He developed a customer segmentation system where AI agents parsed user queries and automated form completion. The system used a vector database, prompt engineering, and multiple specialized agents. He containerized it with FastAPI and worked with engineering to deploy it to Kubernetes.

He also built BaristaBot, a cafe ordering system demonstrating AI agent workflows with LangGraph and FAISS. His AI agent work emphasizes practical system design: handling state management, coordinating multiple agents, and creating deployable systems.