Initial Design

[ericpan64]

Add notes here as things come-up -- current thinking is:

1. Make this a relatively quick demo (Cult of Done)
2. Use LLMs for interactivity -- e.g. custom train Llama or use GPT-4 based on semantic topics (use LLM script to categorize)
3. Help users build a "repository" of: 1) words learned, and 2) articles read
4. Tailor based on difficulty level, e.g. I am at a HSK-1 level and want to "level-up". I want to learn topics that I find interesting, though I want it customized to my current reading level
5. Try to KISS -- e.g. maybe Gradio for parts of it?

... etc. more as it comes up

Close criteria: a design doc with clear actions on what to implement (to get more info, a la Scrappy prototyping!)

[ericpan64]

Some thoughts on stack to use:

• Frontend: Svelte!
• Backend: Actix Web (main + LLM calls as Actix agents) + Python (data processing and cleaning server)
• Database: MongoDB + Postgres (dual usage). Find some tool for managing schemas
• Cache: Redis (if it makes sense)
• LLMs: together.ai (use credits), or local llama deploy
• Deployment: Podman (try it out -- daemonless so more secure!)

Try this out after graduation! Give it like a month, then go again