Local AI Lab 003: Summarizing Qurio Notes with Ollama and n8n

Question

Can a local language model generate useful summaries and learning artifacts from existing Qurio notes?

Hypothesis

A small local model such as qwen2.5:3b should be capable of generating useful summaries and extracting key ideas from technical notes, but output quality will depend heavily on prompt design.

Setup

Hardware

• Dell OptiPlex 3050 Micro

Operating System

• Ubuntu Server 24.04 LTS

Services

• Docker
• Ollama
• n8n

Model

qwen2.5:3b

Steps

Manual Trigger
    ↓
Edit Fields
    ↓
HTTP Request
    ↓
Ollama Generate API
    ↓
Structured Response

1. Created a new n8n workflow called Qurio Content Summarizer.
2. Added a Manual Trigger node.
3. Added an Edit Fields node containing content and prompt.
4. Configured an HTTP Request node to send JSON requests to Ollama.
5. Used the model qwen2.5:3b.
6. Disabled streaming with "stream": false.
7. Sent the note Docker Basics as the input content.
8. Requested four outputs:
   • summary
   • key takeaways
   • action items
   • questions worth exploring

Request Shape

{
  "model": "qwen2.5:3b",
  "prompt": "Summarize the note, extract key takeaways, propose action items, and suggest follow-up questions.",
  "stream": false
}

Result

The workflow successfully generated structured output from an existing Qurio note.

The model was able to:

• produce accurate summaries
• extract the main concepts
• generate useful follow-up questions
• identify plausible next steps

The strongest parts of the output were usually the summary and the follow-up questions.

The weakest part was action items, which sometimes became generic instead of staying tied to the actual learning goals of Qurio.

What Worked

• Local inference worked reliably.
• Docker, Ollama, and n8n integrated successfully.
• The workflow stayed simple enough to understand and iterate on.
• The model accurately identified major concepts from the note.
• Follow-up questions were often useful for deciding what to learn next.

What Failed

• The initial prompt produced action items that were too generic.
• The model lacked context about the purpose and voice of Qurio.
• Output quality shifted noticeably with prompt wording.
• Frontmatter and Markdown structure likely added some unnecessary noise to the input.

Lessons Learned

• Small local models are capable of useful knowledge extraction tasks.
• Prompt design matters more than the infrastructure at this stage.
• The bottleneck is no longer connectivity or deployment.
• The bottleneck is now workflow design and prompt quality.
• Local AI can assist documentation workflows without depending on cloud services.
• This experiment showed that a lightweight local AI stack can turn existing notes into structured knowledge outputs.

Next Experiment

Experiment 004 should test whether Ollama can generate structured Qurio artifacts rather than summaries alone.

A likely workflow is:

Raw Learning Material
    ↓
Ollama
    ↓
Structured Markdown Draft
    ↓
Qurio Note Template

Questions worth exploring next:

• Can a local model generate complete draft notes?
• Can journal entries be partially automated?
• Can experiment templates be generated automatically?
• What is the quality difference between qwen2.5:3b and larger models?

The infrastructure is now proven. The next layer of work is improving prompts and generating more useful artifacts that support the Learn -> Build -> Document -> Improve workflow.