Hi David, Since flink stateful has been inactive for some time[1], I'm not sure whether you could get some help here. From my understanding, your request to a distribute system is event driven processing + actor model. Perhaps you could enrich Ray[2] with continuous message processing to achieve this goal, since Ray already has the ability of object store.
BTW, since LLM is really popular in the current data + AI era, maybe you could also share some insights here after the investigation. [1] https://lists.apache.org/thread/7cr2bgt91ppk6pz8o0nfbd10gs63nz6t [2] https://docs.ray.io/en/latest/ Best, Yun Tang ________________________________ From: Yunfeng Zhou <flink.zhouyunf...@gmail.com> Sent: Monday, May 6, 2024 9:03 To: dev@flink.apache.org <dev@flink.apache.org> Subject: Re: Flink stateful functions and Agentic Architecture Hi David, I'm not very familiar with stateful functions, but I participated in Flink ML, a machine learning infrastructure and algorithm library based on Flink. There we developed functions like iteration based model-training process and hot updating ML model during online prediction. You may check if these functions and their corresponding designs could be of your interest. https://nightlies.apache.org/flink/flink-ml-docs-release-2.3/docs/development/iteration/ https://github.com/apache/flink-ml Best, Yunfeng On Mon, Apr 29, 2024 at 9:35 PM David Carroll <da...@spotter.la.invalid> wrote: > > I am a systems architect developing a POC concept for an AI product using > Agentic Architecture with generative LLMs. It occurred to me that it could > be possible to use Flink stateful functions to provide the event driven > communications and execution environment for agents built with a framework > like langchain or langgraph. Such a system could be multi-tenant and fully > scalable. > > I wonder if anyone has explored using Flink stateful functions i this way > of has suggestion of examples to look at for similar use cases? > -- > David Carroll > Chief of Research & Development > <http://spotter.la/> > da...@spotter.la | 310.569.5103 > <https://time.com/collection/time100-companies-2023/6285153/spotter/> > From TIME. © 2023 TIME USA LLC. All rights reserved. Used under license. > > IMPORTANT: The contents of this email and any attachments are confidential. > They are intended for the named recipient(s) only. If you have received > this email by mistake, please notify the sender immediately and do not > disclose the contents to anyone or make copies thereof.