Automatic Priority and Time Labeling

[0x4007]

This plugin may require vector embeddings of the issue specifications, but it seems to be a responsibility under this plugin for dealing with the labels so I'm filing here.

The system should learn from issue specifications and their labels to predict priority levels and time estimates for newly posted tasks.

We should be able to configure to automatically assign time, priority, and/or price labels.

For example if

autoSetTime: true
autoSetPriority: true
autoSetPrice: false

Then this will allow the team to approve of task pricing before allowing people to self assign.

Precision

This could be a second generation of this capability. It could be especially useful when computing against larger tasks in the weeks or months range.

It would be great to generate custom time labels for added precision. For example, if historical task A took an hour and task B took two hours, and a new task falls in the middle in terms of similarity, it could assign a label of 1.5 hours. We technically could do the same for priority levels, and get precise pricing generated.

I would imagine that the more data the system has the more accurate these labels will be.

[ubiquity-os-beta]

Note

The following contributors may be suitable for this task:

gentlementlegen

78% Match ubiquity-os/plugins-wishlist#6

whilefoo

76% Match ubiquity-os-marketplace/daemon-pricing#45

[0x4007]

This is a big prerequisite towards the future vision of UbiquityOS being able to have a quarterly conversation with leadership and right after to generate hundreds of tasks with pricing instantly.

With task matchmaking and disqualifier it can also assign these and ensure that they are completed.

[0x4007]

@shiv810 rfc on time estimate

[shiv810]

@shiv810 rfc on time estimate

This should take about a week to implement. There are two key components: first, scraping the actual time taken, priority, time label, price assigned, and task specification embedding. Second, using the scraped data, we will develop a machine learning model that takes the embeddings as input and returns all three values: time, priority, and price. This machine learning model would have to something simple. Alternative would to fine tune an openai model, and then use that for inference, but I am not sure if you can use embedding data as a input for fine tune.

[0x4007]

That seems way over engineered to make a custom AI model we just need linear regression. Normalize time to milliseconds. Priority also has a numerical value.