Sean Clayton SeanClay10

Sean Clayton

Welcome to my GitHub!

Academics: Senior Computer Science student at Oregon State University, concentrating on ethical AI implementation, machine learning, and software architecture.
Technical Interests: Agentic AI systems, machine learning, RAG architectures, and scalable software infrastructure.
Professional Goals: Build impactful, privacy-conscious software and AI tools that solve real world problems.
Hobbies: Soccer, hiking, and playing guitar.

Description: An agentic AI system that enables natural language querying of smart building sensor data, allowing facility staff to analyze environmental conditions without writing database queries or performing manual analysis.
Key Features:
- LangGraph-based agent workflow orchestrating LLM reasoning, validation, data retrieval, and analytics execution.
- Structured intent extraction from natural language using a locally hosted LLaMA 3.1 8B model with schema-constrained JSON output.
- Deterministic analytics tools for temporal statistics, spatial comparisons, aggregations, and threshold monitoring across building sensors.
- Interactive Streamlit interface with visualizations, execution trace transparency, and exportable analytics reports.
Technologies Used:
- Python
- LangGraph
- Streamlit
- pandas

Description: A pipeline written in Python that leverages XGBoost and LLMs to extract predator-prey interaction data from a global database of diet surveys, enabling the validation of the fraction of feeding predators.
Key Features:
- Preprocesses ecological text data and applies TF-IDF vectorization for feature extraction.
- Uses XGBoost to classify relevant publications with key predator diet metrics.
- Utilizes local LLMs for deeper extraction and analysis of complex unstructured text.
Technologies Used:
- Python
- XGBoost
- sk-learn

Description: An autonomous AI agent that plays Pokémon Red using computer vision and a locally hosted Vision-Language Model, eliminating API costs while maintaining complex decision-making capabilities.
Key Features:
- Achieves autonomous gameplay through real-time screenshot analysis and strategic decision-making using Qwen3-VL.
- Reduces inference costs from ~$100 per playthrough to $0 by running entirely on local hardware with 4-bit quantization.
- Hybrid perception system combining computer vision, RAM memory hooking, and collision map generation for robust spatial reasoning.
Technologies Used:
- Python
- Qwen3-VL
- PyBoy
- HuggingFace Transformers

Description: A performance optimized poker simulator written in C++ that implements advanced data structures and algorithmic techniques to simulate realistic poker gameplay.
Key Features:
- Multi-threaded functionality for enhanced performance.
- Realistic game mechanics, including betting rounds, hand evaluations, and decision-making logic.
- Scalable design allowing for multiple players and various poker variants.
Technologies Used:
- C++
- Qt
- CMake

Languages: C, C++, Python, JavaScript
Tools & Frameworks: LangChain, React, Node.js, sk-learn, Docker, Kubernetes
Methodologies: Agile Development, Test-Driven Development (TDD), Scrum Methodology

I'm always open to collaborating on exciting projects or discussing ideas. Feel free to reach out: