LLM Chatbot to support sales team
2024 · Kerecis → Coloplast
End-to-end design of a production chatbot. A RAG pipeline over a vector database cut average response time by 90% across the majority of incoming queries.
I build AI systems that operate in high-stakes, human-centered domains — particularly healthcare. My work lives at the intersection of large language models, retrieval-augmented generation, and neural network efficiency, with a constant emphasis on making these systems reliable enough to trust in production.
Before industry, I spent five years at Clemson studying how to make deep networks smaller and faster without sacrificing what makes them useful. Eight publications came out of that work.
End-to-end design of a production chatbot. A RAG pipeline over a vector database cut average response time by 90% across the majority of incoming queries.
Doctoral research on making deep networks dramatically smaller without losing accuracy. Pioneered genetic algorithms for dynamic pruning and quantization of DNNs, achieving significant model acceleration. Published in IEEE ICPR and seven other venues.
Python · PyTorch · TensorFlow · LangChain · XGBoost · AWS (S3, Lambda, SageMaker, CloudFormation) · MLflow · SQL/NoSQL · Node.js · HIPAA/HL7/FHIR · ICD-10
Volunteer with Arlington Neighborhood Village, providing tech support and digital literacy help to older adults in the community — troubleshooting devices, setting up accounts, and making everyday technology feel a little less daunting.
Agentic AI architectures · LLM evaluation and safety · model compression for edge inference · the intersection of clinical workflows and machine intelligence