AI is actively revolutionizing nearly every industry. The non-deterministic nature of building AI and LLM-powered systems creates a unique set of challenges and opportunities, where creative solutions and abstract thought are the tools that allow us to achieve what was previously impossible. I have been working on LLM-based systems since the release of chat-gpt, and have developed a set of tools and experience to address some of the main challenges in the field, including:HallucinationsContext window limitationsConsistency/reliabilityScalabilityTransparencyEvaluabilitySome relevant well-known tech:RAG systemsChain of ThoughtdspyArtifacts (I have had this for a while but called it something else; now it has a name!)Reinforcement LearningSelf-summarizationExperimental Tech:Dynamic Embedding-based Categorizational TreesSelf-Contained Dynamic Agentic LLM systemsRelational Semantic EmbeddingsDynamic, Task-Relevant, Computationally-curated Memory Automatic Software Dev ToolsWe are at a strange point in the AI Revolution, where it is simultaneously clear that:1. LLM-based systems can accomplish a seemingly boundless array of tasks at the proof-of-concept level.2. Making a proof-of-concept that gets 80% of the way to an MVP is easy.3. Optimizing, refining, and scaling an LLM-based system to that 100% (or even 95%) can feel impossibly difficult without making significant compromises to product vision.In my research, I am focusing on building solutions to address this issue, including the experimental tech above, keeping up with cutting-edge research, and developing custom solutions to problems that I face in my projects.Currently Head of AI at Cadenza, developing LLM-powered tools for non-profits to assist with impact reporting, grant writing, internal document search, and more.Owner at Guru Cloud & AI, LLC, where I do contracting, consulting, and R&D for novel approaches to modern LLM-system problems.Previously Head of AI at NewSeat Co, where we worked to develop AI solutions in the recruiting + HR sphere.