Hello community! I'm Hayden Barnes. My background is in systems, automation, DevOps, cloud native, and for the past couple years I have been doing MLOps and DataOps with Determined AI and Pachyderm, respectively. Lately I have been exploring Prefect. For my first project yesterday I put together a rudimentary workflow that clones the Prefect docs, converts and chunks the .mdx files, creates embeddings, deploys a model with embeddings behind an API, and then deploys a very basic web frontend for a chat bot to query the API. The model is not well-tuned (answers can vary wildly, see TODO/known issues in project README), but I wanted to get used to working with

@task

and

@flow

, the Prefect server, web UI, and just working with the orchestrator. I would welcome feedback on my project, particularly how to take more advantage of features of Prefect. For example, I launch a FastAPI process to deploy the API and a Flask process to deploy the web front end using

process

from

multiprocessing

but I think within Prefect it would make more sense to use

.submit()

to run them on task runners. What do you think?

hi @Hayden Barnes - cool! thank you for sharing you may be interested to check out this example where I: • scrape docs • chunk • embed • load to a vectorstore like chroma i use some prefect features like mapping, and caching and stuff like that there's a slackbot here that consumes these embeddings by using function calling to answer questions based on user queries for example

Very interesting @Nate, going to have to check this out and your

raggy

project!

let me know if you have any questions about it or prefect -

raggy

is just all the rag utils that we wanted to move out of marvin a while ago when we made that library more just about structured outputs. I'm probably one of the only users 🙂 but I would love any feedback or ideas for contributions on that

Welcome, Hayden ✨🎊🙌! Thank you for introducing yourself, and for sharing your project! If you'd like to share this project and others you're working on with the community, #CL09KTZPX is waiting for ya