All you need is modern software practices: developing LLM tools the right way
Come hear a tech talk to rival the greats.
What is Tool Calling?
Tool calling enables large language models (LLMs) to interact with external systems, execute tasks, and take dynamic actions. This capability forms the foundation of agent-driven AI.
Developing Tools for LLMs Well
Implementing modern software engineering practices, such as versioning, packaging, and CI/CD, is crucial for developing reliable, scalable, and efficient AI tools.
These practices ensure tools are consistent, easily shared, and well-documented, while CI/CD streamlines testing and deployment to reduce errors and enhance integration. Clear documentation and parameter management further improve usability and teamwork.
AI-focused strategies, like avoiding issues caused by fragile parsing methods in tool definitions, help eliminate inefficiencies and resource wastage. Tailored evaluations based on specific use cases are vital to gauge the effectiveness and versatility of LLM toolsets in various scenarios.
Does your tool come with a pre commit config?
Do you think you can get away without having a make file?
Can you do a Python poetry publish to make your tool consumable?
And what about linting?
We need to bring these standard software practices to our AI functions. We need to treat tools as code not fancy prompts and one-offs in jupyter notebooks.