AI stacks - terrytaylorbonn/auxdrone GitHub Wiki
25.0717 Lab notes (Gdrive), Git
This Wiki section "AI LLM stacks" is a very methodical, structured, and organized approach (WIP) for DEMO'ING, ANALYZING, and DOCUMENTING AI LLM stacks. This is how I work (I am looking for a job).
Section 4 Deep dive is perhaps the core goal of the AI LLM stacks wiki. I put together section 4.0 Deep Dive Concepts after many conversations with Copilot. Copilot's assessment of my approach (see docx #499):
- "🏆 Overall Assessment: This is wiki-ready! It fills a real gap - I haven't seen anything this concise yet comprehensive. The combination of theory + your actual code examples makes it especially valuable. Recommendation: Publish this! It would be extremely helpful to the community."
For this personal sandbox project, the focus is on this wiki and the Gdrive docx's (internal planning docx #499 provides a rough (and possibly outdated) overview). But for a customer project, the wiki would just be an intermediate step in creating the core tech doc deliverables.
- Test websites for exploring (as one does in a sandbox) various website tech, mainly for doc websites (docsX.ziptieai.com).
- Gdrive lab notes (docx files).
This section follows a logical progression from beginner to advanced (how to approach learning AI stacks). Details of each section:
- 0a Demo deployments. The end goal (typically on Render, etc; may also list local model "deployments"). Should motivate the reader.
- 0b Platforms and tools (Git, Render, AWS/GCP, etc). Instead of starting at the tech bottom ("what is a tuple?"), start at the top, learning how to deploy basic demos. This can be difficult because AI LLMs struggle (for now) to cover (accurately) platform and tool details.
- 1 Youtube demos. These Youtube demos provide the details that an LLM may not (especially for platforms and tools).
- 3 Hackathons (ChatGPT "sprints"). My favorite dev method. Just ask the dev LLM chatbot to guide you.
- 4 Deep dive. Embellishing demos from 1 and 3. Become an expert in a few selected topics. See especially 4.0 Concepts.
There are basically 2 types of docs, for the
- 6 LLM/agent
- 5 Human reader
Details:
- 6 LLM/agent docs. The new AI doc paradigm. Delivered to you in-context in your dev environment (just like Amazon delivers to your home).
- Doc content is created specifically to be scraped, repackaged, and presented to the user by AI-based tools (chatbots like ChatGPT).
- In the future, LLM dev tools might just auto-update scraped docs (product dev teams only have to publish the content; the LLMs will automatically take care of the rest). Perhaps websites will be designed for AI (see llms.txt)
- 5 Human reader docs. Traditional "brick and mortar" docs (mostly website based). You visit them to get what you need.
- Note: Seems to me that published docs might start focusing on the prompt chains needed to accomplish tasks. The idea is not to put all the details into a doc, but let the LLM (GPT/CPLT) produce the details (that are customized to the user content).
(notes)
- #1 SS/LIN: GPT is best (like driver assist... GCLI / CPLT get you killed).
- #2 SS/LIN:
- future docs
- add tages to docx files (mainly focus on prompts)
- AI picks up and publishes
- suggest, ask, adjust, step by step (miserable github UI/docs)
- future tools
- sites designed for ai
- ai designed for sites (auto update)
- future docs
- /mnt/c/Users/terry/Downloads$ git clone https://github.com/terrytaylorbonn/auxdrone.wiki.git (to avoid win11 \ error)
- GPT feedback about the above structure.
- extract content and eventually delete A1-3 PRIVATE AI stack deploys and A1-4 PUBLIC AI stack deploys.
- AI stack FE/BE compatibility 25.0416