Todays Crew Project!

General
1

Hi All,
Last night I asked my wife if there was anything that a crew could do for her?
Her response was such a novel suggestion I thought that I would put it here for others to play with:

In our house we have many externall HDDs where box sets and films that I’ve downloaded are stored. Most evenings my wife & I have the ‘what should we watch’ discussion. These discusions can take 30 mins or more given the 1000+ films we have, occassionaly we think of films that we had forgotten about. Basically this nightly ritual is becoming tedious:

From my wife:
Get your crew ‘thing’ to scan the HDDs and create a cataloug of all of the films, box-sets so we don’t have to have these annoying discussions & jumping fom HDD to HDD every night :slight_smile:

Plan:

  1. Create a tool that given a root path can scan a directory system & return filenames of all films found.
  2. Design a crew of Agents/Tasks that will use the tool and do things like, find details of each film online & provide such as: user ratings; summary; key actors; genre, etc

@matt Suggestion: A monthly CrewAI project/task where suggestions are put forward for a CrewAI crew project,/task voted upon, selected. Members can then submit there crew solutions for the project/task for evaluation, comment, & discussion. == Group learning :grinning:

2 Likes
2

Love this idea, let me look into the how’s an stuff

3

I have created a Github Repo for my attempt at a solution to this.

I also have a GitLab account, if any of these learning projects turn into something useful I’ll move the repo to GitLab and set up CI/CD test /build/deployment pipelines to push out Docker container images. If others would like to learn, etc.

If anyone would like to contribute, let me know via PM, or post below.
Code suggestions welcome.

Comments:
Take note of how I instantiate list objects on the crew instance, and pass these into tools for them to work on. I also have tools that a Task/Agent uses to save their output into these crew list instances, and recall info from them via other tools. Still checking, but it looks like this stops reams of Agent/Task output being put into the context, which to me is a good thing. :grinning:

Questions:

  1. Can a ‘tool’ have more than just the _run method? If you look at the tools that I had to make to manage a ‘list’. A CRUD tool with multiple methods would have been better.

ALL comments welcome. **REM: It’s all about learning.

4
1 Like
5

Thanks @matt.

I created the repo to save forum space! Some of the posts were turning into books of code!

I’ve just given it public visibility, I’ll link it to the forum post for reference

6

NOTE how I use the Pydantic model Field descriptions in the prompts.

They are loaded into the crews inputs here

@moto :grinning: