Large language models (LLMs) have transformed the way we interact with and leverage AI for problem-solving. However, as these models grow in capability, the way we manage and represent knowledge must evolve alongside them. My current research is focused on improving context memory, knowledge representation, and reasoning — not by building new models, but by innovating the tools and frameworks that make LLMs more effective in real-world scenarios.
Here’s a brief look at the areas I’m currently exploring:
1. Dynamic and Persistent Context Memory
To enable dynamic, multi-session interactions with LLMs, I’m developing identifiable, referenceable, and addressable data blocks. These blocks provide persistent storage and retrieval mechanisms, ensuring continuity across diverse tasks.
2. Encoding Relationships Between Data Blocks
To build meaningful knowledge structures, I’m designing systems that encode relationships between data blocks. This involves encoding diverse relationships — hierarchical, temporal, causal, and beyond — to create structured, interpretable knowledge graphs that adapt to a wide range of reasoning tasks and real-world applications.
3. Dynamic Typing for Natural Language Prompts
A significant focus of my work is on enabling dynamic typing for data blocks. This allows them to adapt to natural language prompts, making them context-aware and versatile in handling diverse tasks.
4. Minimalist Syntax for Structured Input and Output
I’m experimenting with a custom syntax that minimizes code indentation while retaining readability and clarity for nested data structures. This syntax is tailored for workflows like Chain-of-Thought (CoT) and Tree-of-Thought (ToT) reasoning, enabling structured problem-solving.
5. Integration with RDF and Linked Data Principles
To ensure interoperability, I’m aligning my work with RDF (Resource Description Framework) and Linked Data principles. This integration helps data blocks fit into the broader semantic web ecosystem and allows for sophisticated querying and relationships.
6. Collaborative Document Models
Collaboration is essential for iterative work, and I’m exploring how block-based document editing can support team-based workflows. By enabling granular edits and version control for individual blocks, this approach aims to simplify collaborative knowledge creation.
7. Practical Research for Toolset Development
Ultimately, my goal is to create a practical, user-friendly toolset that bridges the gap between research and application. By focusing on modularity and accessibility, I hope to make these tools valuable to a wide range of users and scenarios.