QS

This page (WIP) describes a set of selected start-to-finish demos for learning the core concepts of AI quickly. If you are new to AI, its best to get hands-on demo experience first (with this QS), and then study the concepts.

TOC

• 2 NN demos. Tiny demos that can also run on a CPU.
• 2b Model demos. Creating your own tiny models.
• 3 External agent demos. Using a model as a helpful assistant (as a standalone system or as a helpful assistant for a target system).
• 3b Workflow framework demos. The workflow platform controls the architecture.
• 3c Enterprise demos (Palantir). The enterprise platform controls almost everything.

Right to left: 2 NNs, 2b Model, 3 External agent, 3b Workflow framework (need to add 3c Enterprise platform)

2 NN demos

Tiny demos that can also run on a CPU (the numbering in the partial list of demos below matches the numbering of the demos).

• 2.1.2 Tiny NN (very basic NN).
• 2.2.1 Tiny CNN (very basic CNN for recognizing digits 0-9).
• 2.3.6.1 Tiny transformer (very basic TF that outputs “hello world…” for input “h”).
• 2.5 Tiny Robotic NN (very basic demo of using NN to avoid obstacles).

NOTE: The core of AI is not rapidly evolving. The simple demos on this site are the best demos if you want to understand the core mechanics of how AI really works. For example, page **2.3.6.1b D5 tiny TF algorithm details (draft, WIP) explains in detail the simplest demo (its not simple) of the LLM TF QKV (context) mechanism.

QKV is still the core of LLM AI (video)

2.1.2 Tiny NN / demo D2ccc

This one simple NN demo will give you an understanding of the core of AI inference (used in NNs, CNNs, LLM TFs, etc). The following diagram shows the core components of the demo:

• Encoder. All NNs only “know” numbers (including the NNs at the heart of ChatGPT, Claude, etc). So if you have a dataset that is not (FP) numbers, it must be encoded (and later decoded).
• NN. The source of “intelligence”. A UFA (universal function approximator). This NN inputs 2 numbers and outputs 2 probabilities. That’s it.
• Decoder. Convert from a probability for inside/outside the circle to a binary state (1/0 = inside/outside circle).

GPT: “This demo teaches tensors, forward(), Linear, ReLU, logits, CrossEntropy, training, inference without overwhelming people.”

2.2 Tiny CNN demo D4 MNIST

Input a 28x28 pixel screenshot of a digit, and the NN outputs “0”, “1”, … “9”. See also 2.2.1b D4 CNN algorithm details.

There are 3 main parts.

• 1 Data input. In this case the data input is from MNIST.
• 2 Feature extraction (nn.Conv2d + nn.MaxPool2d). This is where convolution and pooling are used to turn the original pixel values into “hidden state” values that progressively define higher level features (I typically call these values “pix’s”).
• 3 Final NN (nn.Linear) computes CNN “logits”. This involves
  • 3.1 Linear NN (like for D2ccc) that computes the probabilities of each of 10 possible outputs (digits 0 to 9).
  • 3.2 The most probable output is converted into a label (such as “7”).

CNN demo D4 overview (original ZiptieAI/GPT diagram)

2.3.6.1 Tiny TF demo D5

TF demo D5 overview (original ZiptieAI diagram)

2.5 Tiny Robotic NN (wiki)

You might want to skip this demo. I was going to focus on robotics AI more, but decided not to.

2b Model demos

A model contains the NN/CNN/TF and an “internal” agent (iAgent) that controls and provides access to the NN/CNN/TF. The iAgent implements an API that provides external access to the model.

• 2b.2 Tiny demos. GPT recommends doing demo M01 (save model) because it introduces NN + .pt.
• 2b.2b Building models. GPT recommends M06 HuggingFace publish because suddenly the model becomes portable.
• 2b.3 Modifying models. TODO. GPT recommends a tiny fine-tuning demo. For example: M08 Tiny Fine Tune
  1. Load M01 model.pt, 2. Run inference,
  2. Continue training for 100 epochs on slightly different data
  3. Save model_v2.pt, 5. Compare outputs before/after
• 2b.4 Running models locally. GPT recommends M07 FastAPI. This is where the model becomes a service. But GPT also says that Ollama is a better representative demo for 2b.4 Running models locally than M07 FastAPI. The purpose of 2b.4 is not to teach FastAPI. It’s to teach: “How do I run a model on my own computer?” Most people today will use something like: Ollama, LM Studio, vLLM, llama.cpp rather than writing their own FastAPI server.

3 (External) Agent demos

Any one of the following demos would be a good start. The focus of these demos is how to create an agent that interacts with several external systems using AI as a helpful assistant.

• 3.1 Code-first. I’d recommend one demo from each of these sections.
  • 3.1.1 Agent PAL_core demos (docx #603).
  • 3.1.2 PAL demos (openAI/Gemma) (docx #603).
  • 3.1.3 TF semantic demos (AI) (docx #606).
• 3.2 Frameworks. I dont have examples for these right now. They are not my priority. I have done LangX demos, but the code intensive nature of the LangX is not what I want to focus on right now. I think its not the way for most newcomers to go, and I wonder about the opportunities in the future (I think higher levels frameworks such as n8n and Palantir are the way to go; if not these tools, then other newcomer tools in the near future).
  • OpenAI Agents SDK, Pydantic AI, LangChain, LangGraph, CrewAI, AutoGen
• 3.3 Projects. Practical AI applications that provide useful functionality. Either of the following.
  • 3b.3 (NMAP) Security Assistant. Uses the local PC as the target system. Works outside the target.
  • 3b.4 Ledger. Stores messy human text in an append-only encrypted ledger, then uses AI to search, summarize, and generate new entries without requiring a rigid database schema. The ledger stays immutable. Then the AI/agent produces the current “view”. So nothing is edited in place. The system only appends.
• 3.4 Integration demos. Demonstrations of integrating the central AI agent with existing software and enterprise systems.
  • 3b.12 Odoo). The focus is on adding AI to Odoo.
• 3.5 Lab demos. Just hacking around. Avoid these..

External agent ecosystem

3b Workflow (Framework) demos

• 3b.1 n8n (26.0501). A basic test using Cursor with minimal manual coding.
  • Main system = n8n LOCAL for test data input.
  • 3b Project = Input data from n8n > gpt-40-mini LOCAL > mongoDB > index.html (UI).
  • 3 External agent = cursor_11_llm_canonical.py
  • 2b Model = gpt-40-mini LOCAL.
    Functional diagram
    
    Ecosystem diagram
    

3c Enterprise demos (Palantir)

• PAL-DEMO-2: Palantir AIP “speedrun” (quick start).
  Final app
  

26.0716 (0628, v1 26.0527)