So... Still on the design stage for my thingy. Tokenizer for my language is still incomplete, requirements parser at 0%, and I'm already wondering what I'd do about case generation for unit tests.
I want my AI to be able to vibe-code itself (bootstrapping). And I got an idea!
I need to add a database of knowledge, eg, token expressions, and then other domains.
I think I'm turning this into an actual Expert System 😁
Let's say a user wants to generate a series of English names, both male and female.
All I need to do is create a SQLite table with common names, genders and popularity.
Then I could expand this to other domains of knowledge, like business use cases.
Then, common algorithms like sorting and searching - for Java that's easy, it's just a few imports and calls 😅
What really tickles my curiosity is modifying tests, like "verify that this test checks that (CASE). How? 🤔
Also, I need to add aliases and context memory.
Like, "function testThatWhateverDoesntCrash()" has a temporary alias such as "the test for this thing" that the user mentioned two prompts ago. So prompts need to have implementation-specific context memory.
If the memory is lost, I can ask for clarification. "Sorry, I forgot. When you said "the first test", which function are you referring to?"
I want to see how far I can get making the user think my AI is an LLM. 😁