Ollama Embedding Issue Ops

Automation isn't just for builds and deployments anymore—let’s build something fun with it. In this project, we'll explore how to turn GitHub Issues into a creative workflow using Issue Ops. We'll leverage Ollama to generate text embeddings, visualize them as vector images, and automatically display these images in the issue itself.

---

What We Need

To follow along, start with the Ollama Devcontainer GitHub template, which comes pre-configured with Python and Ollama.
You'll just need to add the following Python libraries to your requirements.txt file:

scikit-learn
numpy
matplotlib

This will set up your environment for embedding text and visualizing vector images.

---

Running Issue Ops

1. Go to this GitHub repo.
2. Select Issues on the top horizontal nav bar.
3. Once the issues dashboard pops up, select the green New Issue button in the upper right corner.
4. Insert any title you want, and add words or phrases into the issue body.

Once an issue is created or edited, a GitHub workflow runs in the background. This workflow extracts the words and phrases from the issue body and then executes the main.py script in the repository. The Python script embeds each word, plots them on a graph using Matplotlib, and outputs the result as a .png file.

---

Generating Word Vectors — main.py

The main engine of this project is main.py.
The script connects to an Ollama model (nomic-embed-text:v1.5) to generate embeddings. These embeddings allow us to understand how words relate to one another in a mathematical space.

Setting Up Environment

Running this in a devcontainer using:

• .devcontainer/devcontainer.json
• .devcontainer/docker-compose.yml

…will create an Ollama server that can be accessed from http://ollama:11434 and a Python workspace to develop in.

Environment Variables

• OLLAMA_HOST: The URL to the Ollama server.
  When running inside the devcontainer, an Ollama server will be available at: http://ollama:11434

• MODEL: The model used to embed text.
  Currently set to: "nomic-embed-text:v1.5".
  Other available models can be found in the Ollama Library.

Loading Inputs

If running this outside of the workflow, you’ll need to create a words.txt file to add your words/phrases.
These should be written on new lines and prefixed with -.

Example:

- king
- queen
- princess
- prince

What main.py Does

1. Generating Embeddings

Each word is sent to the Ollama model, which returns a high-dimensional embedding vector. Think of this as a fingerprint of the word’s meaning, with hundreds of numbers encoding semantic context.

2. Dimensionality Reduction with PCA

Since these embeddings are too large to visualize directly, the script applies Principal Component Analysis (PCA). PCA compresses the data down to three dimensions while preserving as much of the structure as possible, making it possible to plot the relationships in a 3D space.

3. Plotting in 3D

Using Matplotlib, the script creates a 3D scatter plot.
Each point represents a word, and its location reflects semantic similarity—words with closer meanings appear closer together. Labels are added with coordinates for clarity, and the visualization is saved as a high-resolution PNG file.

---

Issue Ops Workflow — issue-ops.yml

Workflow Name & Trigger

• Name: Run Vector Image Create
• Trigger: Runs when a GitHub issue is opened or edited:

on:
  issues:
    types: [opened, edited]

This means anytime someone creates or modifies an issue, this workflow will automatically kick off.

Important Steps of Workflow

Save Issue Body to File

Takes the text from the issue body and writes it into a file called words.txt.
This will be the input for the Python script.

Run Script & Generate Image

Runs the main.py script and saves the image created to the img folder.

Commit Image Back to Repo

Commits and pushes the image created by main.py into the repository.

Comment on the Issue

Uses peter-evans/create-or-update-comment@v3 to post a comment on the original issue with the image embedded.

---

Conclusion

This project shows how automation can be used not just for routine DevOps tasks, but also for creative and exploratory workflows. By combining GitHub Issue Ops, Ollama embeddings, and visualization with Python, we’ve built a system that transforms plain text into meaningful, interactive images. Each issue becomes more than just a record of words—it becomes a visual map of relationships and meaning.

The best part is that everything runs automatically in the background. Contributors simply open or edit an issue, and the workflow takes care of embedding, plotting, and posting the visualization back into the conversation.

It’s a fun example of how we can blend AI, automation, and collaboration platforms like GitHub to make everyday tools more engaging.

Whether you want to explore semantic relationships between words, experiment with embeddings, or just add some creative flair to your GitHub projects, this pattern can serve as a starting point for countless other ideas. Automation doesn’t have to be boring—it can be powerful, educational, and even fun.