Autonomous Product Team

using autogen to set up a multi-agent team of specialized LLMs which mimic a product development and research team ...

steps to set up the jupyter notebooks locally in a virtual environment:

• install python
• clone this repo and change directory to come into it
• initiate a python virtual environment:
  • python -m venv venv
• activate the venv:
  • . venv/bin/activate
  • prompt should show (venv) ...
• install the required python libraries
  • pip install -r requirements.txt
• (optional) set up jupyter extensions:
  • jupyter contrib nbextension install --user
  • dark mode:
    • jt -t onedork -fs 95 -altp -tfs 11 -nfs 115 -cellw 88% -T
  • reset dark mode:
    • jt -r
  • the article is here
  • the documentation for the extensions here
• run the notebooks:
  • cd notebook # to save the notebooks here
  • jupyter notebook

steps to download opensource LLMs to your laptop

• tbd (use LM Studio)

steps to run these LLMs behind an (open-AI compatible) API

• tbd use LM Studio
• need to figure out how to run multiple LLMs together, local-ai didnt work, LM Studio doesn't allow it natively

steps to kick off docker container for code execution

• this is where we will get the coder LLM (our AI software engineer) to run the code.
• I already have the Dockerfile in the root directory from the CLI run:
• now we build the image (this step takes a while)
  • docker build -t autogen-project .

(venv) tcm autonomous-product-team$ docker build -t autogen-project .
[+] Building 125.9s (5/9)                                                                               docker:desktop-linux
 => [internal] load build definition from Dockerfile                                                                    0.0s
 => => transferring dockerfile: 341B                                                                                    0.0s
 => [internal] load .dockerignore                                                                                       0.0s
 => => transferring context: 2B                                                                                         0.0s
 => [internal] load metadata for docker.io/library/python:3.9.18-slim                                                   3.6s
 => [auth] library/python:pull token for registry-1.docker.io                                                           0.0s
 => [1/4] FROM docker.io/library/python:3.9.18-slim@sha256:96be08c44307e781fd9ce8e05b49c969b4cb902ec23594f904739c58d  122.3s
 => => resolve docker.io/library/python:3.9.18-slim@sha256:96be08c44307e781fd9ce8e05b49c969b4cb902ec23594f904739c58da3  0.0s
 => =>.....
 ```

 - now we run the image:
 	- `docker run -it -p 3010:3010 --rm --network host autogen-project`
 	- -it allows you to interact with the container
 	- --rm removes the container when it stops
 	- --network host allows the container to connect to localhost where I'm running the actual LLM (much faster on Apple metal than a tiny container)