PoC: Leveraging AI for Active Development

As a maintainer of the Source2adoc application, I want to explore the potential of AI-driven feature requests and development to ensure continuous improvement and innovation within the app. This initiative is aimed at leveraging cutting-edge technologies to enhance user experience and maintain the app's relevance in a rapidly evolving landscape.

As a maintainer, I seek to integrate AI into the development process of source2adoc to streamline feature requests and further advancements. By harnessing the power of AI, I aim to accelerate the pace of development while ensuring that user feedback is promptly addressed and incorporated into the app. Using AI should ensure the App remains under active development while minimizing manual effort.

To achieve this, I envision implementing a Proof of Concept (PoC) that demonstrates the capabilities of AI in analyzing user suggestions, prioritizing feature requests, and generating insights for future development roadmaps. This PoC will serve as a foundation for establishing an AI-driven development pipeline that empowers the source2adoc team to adapt swiftly to evolving user needs and technological trends.

By embracing AI-driven development, we aim to foster a culture of innovation, agility, and responsiveness within the source2adoc community, ensuring that the app remains at the forefront of document conversion technology. This PoC represents the first step towards a future where AI plays a central role in shaping the evolution of source2adoc, empowering us to deliver unparalleled value to our users.

Possible functionalities of the AI-enabled development process include (as suggested by ChatGPT)

Feature Request Analysis: The AI system will analyze incoming feature requests from users, categorize them based on relevance and feasibility, and provide insights into the most requested features.
Prioritization: Leveraging machine learning algorithms, the system will prioritize feature requests based on various factors such as user feedback, market trends, and technical feasibility, ensuring that development efforts are focused on high-impact features.
Trend Identification: By analyzing user interactions and industry trends, the AI system will identify emerging needs and technological advancements, enabling proactive development of features to stay ahead of the curve.
Automated Development Support: The AI system will assist developers by generating code snippets, suggesting design patterns, and providing recommendations for implementation, thereby accelerating the development process and reducing manual effort.
Continuous Learning: Through iterative feedback loops, the AI system will continuously learn from user interactions and development outcomes, refining its algorithms to better serve the evolving needs of Source2adoc users.

My ideas about possible features

Somehow use an AI to generate user stories (based on collected data - sommerfeld-io/source2adoc#13)
- Force AI to write user stories. AI should think about personas (= adopt a certain point of view when writing stories)
- Think about how to handle the user stories ... add label and establish some sort of backlog refinement for all these AI-driven user stories.
- Maybe add a label "from-ai" or similar and add a column to my GitHub project "ready-for-refinement" or similar. Think about a "ready-for-development" column as well.
- "ready-for-refinement" could be interesting anyway - even for stories written by me (because I often add a remark that there is work to be done instead of the story content) ... but when I do this for every story, "Backlog", "Ready for Development" and "Next" (instead of "Selected") would suffice.
- Tasks are always "Ready for Development", user stories are not
Prefer GitHub Copilot as AI because all these user stories and all the data and input passt to the AI will enhance the learning of GitHub Copilot for this app.

See

https://github.com/cli/cli/actions/runs/9086700583/workflow

Google Gemini Free Tier

https://ai.google.dev/docs

Request

curl \
  -H 'Content-Type: application/json' \
  -d '{"contents":[{"parts":[{"text":"Explain how AI works"}]}]}' \
  -X POST 'https://generativelanguage.googleapis.com/v1beta/models/gemini-1.5-flash-latest:generateContent?key=<API_TOKEN>'

Response

{
  "error": {
    "code": 400,
    "message": "Gemini API free tier is not available in your country. Please enable billing on your project in Google AI Studio.",
    "status": "FAILED_PRECONDITION"
  }
}

Relates to

ChatGPT infos

There are several AI tools and techniques you can use to enhance your software development:

Code Review and Quality Analysis: Tools like Codacy or DeepSource use AI to analyze your code for potential issues, ensuring higher code quality and consistency.

Automated Testing: Tools like Testim or Mabl leverage AI to create and run automated tests, making it easier to catch bugs early.

Code Generation: Beyond GitHub Copilot, you can explore tools like TabNine or Kite, which offer AI-driven code completion and suggestions across multiple languages.

Natural Language Processing: You can integrate NLP models (like those provided by Hugging Face) to build features such as chatbots, text analysis, or sentiment analysis within your applications.

AI-powered Debugging: Tools like Sentry can help you identify, diagnose, and fix errors in your code by analyzing patterns and common issues.

Project Management: AI tools like ZenHub or Clubhouse can help in managing your projects more efficiently by predicting timelines, prioritizing tasks, and providing insights based on your project data.

Exploring these tools can help you streamline various aspects of your software development process. Which of these sounds interesting to you?

:zap: The issue with automated Code Reviews (due to available services)

Using ChatGPT with GitHub Actions like code-review-gpt or using the OpenAI API directly consumes credits for ChatGPT. I already pay for GitHub Copilot. I do not want to pay for a second AI solution. Plus (I assume) it is only a matter of time until Copilot is deeply integrated into GitHub and allows working with PRs and Issues and everything.

Additional non-AI logic

:warning: for Backlog refinement, not Code Reviews

Each story gets a business value and sorry points (or similar) in a dedicated field (T-shirt sizes for both)
Score for each T-Short size.
- High score for high business value
- High score for low story points
Calc: business_value x story_points = ranking (which todo should come next? get some sort of prio / ranking)
Run as GitHub actions workflow … assign and remove labels? Just 8 items as „next“ or similar???
- Or Better: somehow visualize in a Board / Table based on the same metrics? So the stories that should be implemented next rise to the top of the view.

:thought_balloon: It might be better to not just prioritize issues based on score but to use an AI solution right from the start where the score is combined with some AI recommendations.

GitHub Copilot Chat

You can store business value and story points for each user story by using custom fields in GitHub Issues. Here’s how you can do it:

Custom Labels: Use labels to represent business value and story points. For example, you can create labels like business-value: high, story-points: 5.

Issue Templates: Create issue templates that include fields for business value and story points.

Querying: Use GitHub’s search and filtering capabilities to filter issues based on these labels.

For instance, to filter issues with high business value, you can search for label:"business-value: high".

Would you like more detailed steps or help with setting up specific filters?

When you use issue templates in GitHub, the inputs are stored as a single text block in the resulting issue. However, you can still make the fields queryable by using labels or structured text within the issue.

Approaches:

Labels:
- Use labels to tag issues with specific story points and business values.
- Example: label:"story-points: 5" and label:"business-value: high".
Structured Text:
- Include structured text in the issue body and use search queries to filter.
- Example: Use a consistent format like Story Points: 5 and Business Value: High.

To filter issues based on these fields, you can use GitHub's search functionality with appropriate queries. For example:

To find issues with 5 story points: is:issue "Story Points: 5"
To find issues with high business value: is:issue "Business Value: High"

Would you like to see an example of an issue template structure for this?

Example Issue Template

---
name: User Story
about: Create a new user story
title: "[User Story] "
labels: user-story
assignees: ''

---

**Story Points:** 

**Business Value:** 

**Description**

Provide a detailed description of the user story.

Querying Example

To filter issues with 5 story points:
```
is:issue "Story Points: 5"
```
To filter issues with high business value:
```
is:issue "Business Value: High"
```

Using labels to tag issues can make it easier to filter and manage them.

Label Naming

"needs triage": This is a common term used to indicate that an issue needs to be reviewed and prioritized. It is widely understood in agile and software development practices.
"needs refinement": This also makes sense and may be more specific in indicating that the issue needs further details or planning.

Both are correct, but "needs triage" is more commonly recognized in a broader context.

Marking Story Points and Business Value

Using Labels:
- Pros:
- Easy to filter and search.
- Can quickly see at a glance.
- Cons:
- May clutter the label list if you have many different values.
As Part of the Issue Description:
- Pros:
- Keeps the label list clean.
- Allows for more detailed information.
- Cons:
- Requires structured search queries to filter.
- Not as immediately visible or standardized.

Recommendation

Story Points: Use labels (e.g., story-points: 5). This allows you to quickly filter and see the complexity of issues at a glance.
Business Value: Use part of the issue description. This provides room for more nuanced explanations and prioritization criteria.

Would you like specific examples or further assistance with setting up labels or issue templates?

needs-triage vs needs-refinement

Yes, that distinction makes sense:

needs-triage: Indicates that the issue is ready for review and prioritization.
needs-refinement: Indicates that the issue needs more details before it can be estimated or reviewed.

Using both labels can help clarify the stages of issue processing in your workflow.

Google Gemini Free Tier

https://ai.google.dev/docs

Request

curl \
  -H 'Content-Type: application/json' \
  -d '{"contents":[{"parts":[{"text":"Explain how AI works"}]}]}' \
  -X POST 'https://generativelanguage.googleapis.com/v1beta/models/gemini-1.5-flash-latest:generateContent?key=<API_TOKEN>'

Response

{
  "error": {
    "code": 400,
    "message": "Gemini API free tier is not available in your country. Please enable billing on your project in Google AI Studio.",
    "status": "FAILED_PRECONDITION"
  }
}

sommerfeld-io / .github