code docs 4 ai

problem:

development documentation typically has a good logical explanation on how to use the product and one or a small number of examples. this is good for humans and search engines but not for LLMs. today people code just by asking LLM to write code so the documentation is invisible. even adding the documentation doesn’t helm since LLMs can’t follow logical chains

our solution:

1. scrape documentation
2. analyze documentation
3. analyze how to make a good example (purpose, core concepts, steps, libraries used)
4. analyze how to make diversified use-cases
5. generate code examples
6. run the code examples and track the errors
7. human feed back – correct the errors
8. save the examples to be used in RAG or post them on the web for pretraining

technologies

copilot for assisting human feedback and UI
google ai for LLM calls and grounding to create higher quality examples
@weave.op for tracing
restuck for control flow, tracing, debugging