Phase 6 — Student Handouts

Session 81 · Week 21 · Handout W

What AI actually is — concepts and honest assessment

Name

Date

Fill in the definition column from the session. Then complete the honest assessment and reflection at the bottom. No hype in either direction — accurate expectations are more useful than excitement or fear.

Core concepts — fill in the definitions

Term	What it actually means	Why it matters for running locally
Token
Parameter
Training
Inference
Context window
Quantization (Q4)
Training cutoff

Models are good at

· Explaining concepts

· Summarizing text

·

Models are NOT good at

· Verifying current facts

· Precise arithmetic

·

Honest assessment

Why does running a model locally (on your own machine) matter, specifically for you and for privacy?

A language model is described as "a mathematical function that predicts the next token." What does this explain about why models sometimes give confident but wrong answers?

Complete: "Before this session I thought AI was ___. Now I understand it is ___."

Sessions 82–84 · Week 21 · Handout X

Ollama — installation, models, and API reference

Name

Date

CLI commands

ollama pull modeldownload a model

ollama run modelinteractive chat

ollama listlist installed models

ollama rm modelremove a model

ollama --versionshow version

systemctl status ollamaservice status

Model recommendations by RAM

Free RAM	Recommended model	Size
4 GB	gemma2:2b	~1.6 GB
6 GB	llama3.2:3b	~2.0 GB
8 GB	llama3.2:3b	~2.0 GB
12+ GB	gemma2:9b	~5.4 GB

Check available RAM: free -h

API: /api/generate

curl http://localhost:11434/api/generate \
  -d '{
    "model": "llama3.2:3b",
    "prompt": "your prompt here",
    "stream": false
  }'

Extract response: pipe to | jq -r '.response'

API: /api/chat (multi-turn)

curl http://localhost:11434/api/chat \
  -d '{
    "model": "llama3.2:3b",
    "stream": false,
    "messages": [
      {"role":"system","content":"..."},
      {"role":"user","content":"..."}
    ]
  }'

Exercises

1.Install Ollama. What version is installed? Which model did you pull and how large is it on disk?

Version

Model pulled

Disk size

2.Ask the model something it should struggle with — a recent event or large calculation. What happened?

Prompt used

What happened

3.Send a prompt via the API with curl + jq. What fields does the full JSON response contain?

Command used

JSON fields

4.Test the stateless API: ask a question, then ask a follow-up WITHOUT including the first exchange. Does the model answer correctly?

First question

Follow-up without context

Result + why

5.Monitor htop while running the model. What CPU and RAM usage do you observe during inference?

RAM used

CPU usage

Tokens/sec

Session 85 · Week 22 · Handout Y

Script that talks to Ollama — template and build guide

Name

Date

The template below is the foundation for any script that uses the Ollama API. Fill in the blanks, then use it to build your chosen script. All three script options use this same query_model function.

The query_model template — fill in the blanks

~/scripts/query-template.sh — fill in the blanks in blue

#!/bin/bash

set -euo pipefail

# Model to use — override with: OLLAMA_MODEL=gemma2:2b ./script.sh

MODEL="${OLLAMA_MODEL:-llama3.2:3b}"

API="http://localhost:11434/api/generate"

function query_model() {

local prompt="$1"

local system="${2:-}"

local payload

payload=$(jq -n \

--arg model "$MODEL" \

--arg prompt "$prompt" \

--arg system "$system" \

'{model:$model,prompt:$prompt,system:$system,stream:false}')

curl -s "$API" -d "$payload" | jq -r '.response'

}

# Check Ollama is running

if ! curl -s http://localhost:11434 >/dev/null 2>&1; then

echo "Error: Ollama is not running. Start it with: systemctl start ollama" >&2

exit 1

fi

# Your code below — use query_model "prompt" "system_prompt"

Choose your script and complete the checklist when done

Option A — explain.sh — reads a file and explains what it does

Takes a file as $1. Reads its contents. Sends to the model with system prompt "You are a concise Linux expert."

Handles missing file argument (usage message + exit 1)

Handles file that does not exist

Limits input to first 100 lines (head -100) to avoid context overflow

Correctly quotes file contents for jq using --arg

Tested on: a bash script, a config file, a log file

Option B — error-explain.sh — explains errors from a log file

Reads last N lines of a log file. Sends to model asking for plain-language explanation of errors.

Accepts log file as argument or reads from stdin

Defaults to last 50 lines; accepts -n flag for custom count

System prompt instructs model to focus on errors specifically

Tested on /var/log/syslog and /var/log/auth.log

Option C — commit-message.sh — generates git commit messages

Runs git diff --staged. Sends the diff to the model. Asks for a commit message in "type: description" format.

Checks that it is inside a git repository (git rev-parse)

Handles the case where nothing is staged (no diff output)

System prompt specifies commit message format

Tested on a real staged change

Final verification — all scripts

set -euo pipefail is present

Ollama running check is at the top

jq -n --arg used for JSON construction (safe quoting)

Script is in ~/scripts, executable, tested with bash -x

query_model function is in ~/.bashrc for interactive use

Session 87 + S96 · Week 23/24 · Handout Z

Final project design + Phase 6 vocabulary checklist

Name

Date

Complete the design document in session 87 before starting to build. The checklist is completed in session 96.

Problem statement

what problem does this solve?

What it does

user does X → tool does Y → output is Z

Course skills used

list phases and sessions

Phase 1 (sessions ___):

Phase 2 (sessions ___):

Phase 3 (sessions ___):

Phase 4 (sessions ___):

Phase 5 (sessions ___):

Phase 6 (sessions ___):

Success criteria

3–5 specific, testable criteria

Build plan

one goal per sprint session

Sprint 1

Sprint 2

Sprint 3

Refinement

Phase 6 — AI, Ollama, and final project vocabulary

I can explain this

I need to review

LLM

token

parameter

training vs inference

context window

quantization (Q4)

training cutoff

Ollama

ollama pull / run

port 11434

/api/generate

/api/chat

system prompt

stream: false

jq -n --arg

keep_alive parameter

cold vs warm start

stateless API

Final course reflection (Session 96)

1. What concept from this entire course changed how you think about computers?

2. Describe your final project in one sentence. Then: what was the hardest technical problem you solved?

3. Complete: "One thing from this course I will apply outside of computing is ___."

4. What is the first thing you will learn or build after this course ends?