MicroProfile AI

MicroProfile AI minutes

16th May 2025

Video Recording

Contributors [each attendee must add themself to minutes]

  • Emily Jiang (IBM)
  • Buhake Sindi (Individual)
  • Don Bourne (IBM)
  • Max Rydahl Andersen (Red Hat)
  • Phil Chung (Oracle)
  • Arjav Desai (Oracle)

Minutes:

  • The current progress of move code from smallrye-llm to langchain4j-cdi
  • Buhake to create a PR from the langchain4j-cdi branch against the langchainej-cdi
  • The project should target MP 7.0 and Jakarta EE 10
  • This basically means CDI api should be 4.0 and Weld impl should be based on that only.
  • The project should not depend on smallrye
  • The pom.xml should be updated to pull in the current version of the specs or impls

9th May 2025

Video Recording

Contributors [each attendee must add themself to minutes]

  • Emily Jiang (IBM)
  • Ed Burns (MSFT)
  • Phil Chung (Oracle)
  • Arjav Desai (Oracle)
  • Don Bourne (IBM)

Minutes:

2nd May 2025

Video Recording

Contributors [each attendee must add themself to minutes]

  • Emily Jiang (IBM)
  • Don Bourne (IBM)
  • Max Rydahl Andersen (Red Hat)
  • Arjav Desai (Oracle)
  • Phil Chung (Oracle)
  • Luis Neto (Payara)
  • Buhake Sindi (Individual)

Minutes:

  • Langchain4j repo structure
  • Instead of using MicroProfile or Jakarta, Max suggested creating a repo called langchain4j.cdi. Using MicroProfile and Jakarta EE, it could be confusing as people might think this is a standard.
  • Do we create an umbrella repo to contain multiple modules including Fault Tolerance, Telemetry, etc?
  • langchain4j-microprofile-jakarta=>langchain4j-cdi (2025-05-09: Microsoft is happy to see this)
  • langchain4j-microprofile.=>langchain4j-*
  • lanchain4j-cdi-build-compatible-ext=>langchain4j-build-compatible-ext
  • lanchain4j-cdi-portable-ext=>langchain4j-portable-ext

22nd April 2025

Video Recording

Contributors [each attendee must add themself to minutes]

  • Emily Jiang (IBM) - only can stay for a few minutes
  • Don Bourne (IBM)
  • Clement Escoffier (Red Hat)
  • Luís Neto (Payara)
  • Arjav Desai (Oracle)
  • Phil Chung (Oracle)
  • Buhake Sindi (Individual)

Minutes:

  • Buhake has updated https://github.com/smallrye/smallrye-llm/tree/langchain4j-microprofile-jakarta to include last week’s GAV changes. It's not pushed to langchain4j repo since we can’t make a release from that until we finalize the GAVs. This can take time depending on how Microprofile hosting in Jakarta project discussion progressed.
  • How do we make progress and not be blocked while the process is ongoing?
  • Can we push to langchain4j repo but not make a release so that we can continue work there?
  • Emily: Push to the new repo with a PR
  • Alternate is to continue in smallrye repo.
  • What do we want to work on next? Buhake suggested MCP support, maybe.
  • Emily: MCP is a good direction
  • Agree on a new meeting slot:
  • Use the doodle from Ed Burns to vote ( discussion here)

  • Langchain4J 1.0 expected in May
  • CR 1 next week
  • MCP
  • New version of the protocol
  • SSE transport is deprecated, replaced with the streamable HTTP
  • Declarative MCP - MCP annotation, Red Hat would like to push them to Anthropic as an official SDK
  • Annotations are a bit too dependent on CDI right now
  • No capabilities extraction yet
  • A2A and ADK
  • Provide a higher level of abstraction to compose agents (which then use A2A to interact)

15th April 2025

Video Recording

Contributors [each attendee must add themself to minutes]

  • Arjav Desai (Oracle)
  • Luís Neto (Payara)
  • Buhake Sindi (Individual)
  • Ed Burns (MSFT)
  • John Clingan (Red Hat)
  • Jan Westerkamp (iJUG)

Minutes:

  • Buhake has updated https://github.com/smallrye/smallrye-llm/tree/langchain4j-microprofile-jakarta to include last week’s GAV changes. It's not pushed to langchain4j repo since we can’t make a release from that until we finalize the GAVs. This can take time depending on how Microprofile hosting in Jakarta project discussion progressed.
  • How do we make progress and not be blocked while the process is ongoing?
  • Can we push to langchain4j repo but not make a release so that we can continue work there?
  • Alternate is to continue in smallrye repo.
  • What do we want to work on next? Buhake suggested MCP support, maybe.

8th April 2025

Video Recording

Contributors [each attendee must add themself to minutes]

  • Emily Jiang (IBM)
  • Clement Escoffier (Red Hat)
  • Don Bourne (IBM)
  • Arjav Desai (Oracle)
  • Luís Neto (Payara)
  • Buhake Sindi (Individual)

Minutes:

  • Group ID / Artifact id/
  • Parent
  • Group id: dev.langchain4j.microprofile
  • Artifact id: langchain4j-microprofile-parent
  • Core
  • Artifact id: langchain4j-microprofile-core
  • Config
  • Artifact Id: langchain4j-microprofile-config
  • Fault tolerance
  • Artifact Id: langchain4j-microprofile-fault-tolerance
  • Telemetry
  • Artifact Id: langchain4j-microprofile-telemetry
  • Portable extensions
  • Artifact Id: langchain4j-cdi-portable-ext
  • Build compatible extensions
  • Artifact Id: langchain4j-cdi-build-compatible-ext

Action: Buhake and Emmanuel will use the above ids for their PRs, Before the release, we will decide the final ids.

31st March 2025

Video Recording

Contributors [each attendee must add themself to minutes]

  • Emmanuel Hugonnet (Red Hat)
  • Phil Chung (Oracle)
  • Buhake Sindi (Individual)
  • John Clingan (Red Hat)

Minutes:

  • Emmanuel suggested that we can provide examples (similar to Langchain4J Spring Boot example) similarly for SmallRye-LLM.
  • Possible concern: Helidon 4.2?
  • Integration with Helidon Fault Tolerance (package private interceptor binding)
  • Buhake suggests possible repo name change:
  • Project groupId: dev.langchain4j.microprofile.jakarta
  • Artifact ID: langchain4j-cdi
  • So, for MP FT: langchain4j-cdi-microprofile-fault-tolerance
  • Clement suggested that CDI is a well-known name of Jakarta so rather use that and it’s easier for developers to understand.
  • Once the groupId is defined on Maven Central, it’s difficult to change (also poses security concerns)
  • More discussion on MP WorkingGroup Slack group.: https://microprofile-wg.slack.com/archives/C07HLJBQ36Y/p1743330727930879
  • Possibility to put this on pause while the discussion of project name is underway and also the possibility of merging MP into Jakarta EE?

24th March 2025

Video Recording

Contributors [each attendee must add themself to minutes]

  • Emily Jiang (IBM)
  • Emmanuel Hugonnet (Red Hat)
  • Don Bourne (IBM)
  • Phil Chung (Oracle)
  • Buhake Sindi (Individual)
  • Luís Neto (Payara)

Minutes:

  • Our repo has been created here
  • Populate the repo
  • Group ID dev.langchain4j.microprofile.jakarta
  • Artifact ID langchain4j-microprofile-jakarta-parent
  • Update the cdi integration
  • Langchain4j-cdi-buildcompatible-extension
  • Langchain4j-cdi-portable-extension
  • Revisit the name at a later time
  • Build infrastructure - deploy and release to maven central
  • Emmanuel and Clement would be able to get the infrastructure done with the help from Dmytro Liubarskyi
  • Emmanuel will set up OSSH username and password for the Github actions
  • New meeting time based on this doodle poll

17th March 2025

Video Recording

Contributors [each attendee must add themself to minutes]

  • Emily Jiang (IBM)
  • Arjav Desai (Oracle) - arjav.desai@oracle.com
  • John Clingan (Red Hat)
  • Don Bourne (IBM)
  • Buhake Sindi (Individual)
  • Emmanuel Hugonnet (Red Hat)
  • Phil Chung (Oracle) - phil.chung@oracle.com
  • Luís Neto (Payara)

Minutes:

  • In the current work in progress branch, we need to fix copyrights and smallrye parent/testing references.
  • Release process: need access to publish to maven central esp. for dev.langchain group.
  • Demo code: https://github.com/ehsavoie/car-booking/

10th March 2025

Video Recording

Contributors [each attendee must add themself to minutes]

  • Emily Jiang (IBM)
  • Arjav Desai (Oracle)
  • Jan Westerkamp (iJUG)
  • Luís Neto (Payara)
  • Phil Chung (Oracle)
  • John Clingan (Red Hat)
  • Ed Burns (MSFT)
  • Emmanuel Hugonnet (Red Hat)

Minutes:

  • Repo will be created this week.

  • Agentic AI

3rd March 2025

Video Recording

Contributors [each attendee must add themself to minutes]

  • Arjav Desai (Oracle)
  • Emily Jiang (IBM)
  • Emmanuel Hugonnet (Red Hat)
  • Ed Burns (MSFT)
  • Yann Blazart
  • Phil Chung (Oracle)
  • Buhake Sindi (Individual)
  • Luís Neto (Payara)
  • Alfonso Valdez (Payara)
  • John Clingan (Red Hat)

Minutes:

  • The repo name langchain4j-microprofile-jakarta is the winning vote.

  • Agentic AI

24th Feb 2025

Video Recording

Contributors [each attendee must add themself to minutes]

  • Emily Jiang (IBM)
  • Emmanuel Hugonnet (Red Hat)
  • Buhake Sindi (Individual)
  • Ed Burns (MSFT)
  • Phil Chung (Oracle)
  • Arjav Desai (Oracle)
  • Ryan Cuprak

Minutes:

  • Maybe a clone and put smallrye-llm in maintenance model.
  • Proposal: Create a new branch forLangchain4j
  • Rename the package names: replace io.smallrye.llm with dev.reponame while the reponame could be either langchain4j-microprofile-jakarta or langchain4j-jakarta-microprofile (pending vote)
  • Then fork the branch to contribute back to langchain4j repo
  • Agentic AI

17th Feb 2025

Video Recording

Contributors [each attendee must add themself to minutes]

  • Emily Jiang (IBM)
  • Emmanuel Hugonnet (Red Hat)
  • Yann Blazart
  • John Clingan (Red Hat)
  • Buhake Sindi (Individual)
  • Victor Lu

Minutes:

  • Contribute the integration of langchain4j to langchain4j repo
  • Finalise the repo name: langchain4j-microprofile-jakarta
  • consensus: John and Emily prefer langchain4j-microprofile-jakarta; Yann and Buhake prefer langchain4j-jakarta-microprofile
  • Emmanuel, Victor don’t have preference.
  • [Yann]: I prefer  langchain4j-microprofile or  langchain4j-jakarta-microprofile - sorry, Zoom does not let me in to join the conversation unfortunately
  • John: If the argument is that we put jakarta first due to name recognition or CDI integration (CDI is a Jakarta project), then all new MP specs will have to put Jakarta first. The AI effort began in MicroProfile and MicroProfile should remain first, or only, in the name. IMHO, we are already conceding ground by adding Jakarta. Even when MicroProfile AI started, Jakarta folks agreed that AI should be done in MicroProfile due to it being "early" in AI's evolution.
  • The repo name will be determined via a vote among committers via MicroProfile googlegroup and microprofile-wg.
  • A repo move plan
  • Maybe a clone and put smallrye-llm in maintenance model.
  • AI Agent? 

10th Feb 2025

Video Recording

Contributors [each attendee must add themself to minutes]

  • Emily Jiang (IBM)
  • John Clingan (Red Hat)
  • Ed Burns (MSFT)
  • Emmanuel Hugonnet (Red Hat)
  • Jan Westerkamp (iJUG)
  • Buhake Sindi (Individual)
  • Phil Chung (Oracle)
  • Arjav Desai (Oracle)

Minutes:

  • Contribute the integration of langchain4j to langchain4j repo
  • Name the repo to be langchain4j-microprofile-jakarta 
  • Check with Eclipse Foundation to see whether it is okay to use the name of MicroProfile and Jakarta. Emily to check with Eclipse foundation
  • If there is no concerns, I will contact langchain4j community to have this repo created
  • We talked about how microprofile only/specific integration or features should be added in a new project. The idea is to use a sub-directory/module for that and not have it in core/main so that jakarta projects can work without microprofile dependencies.
  • One of the issues may be microprofile config since it's used in core/main. Emily mentioned that config specification is being discussed in both communities.

3rd Feb 2025

Video Recording

Contributors [each attendee must add themself to minutes]

  • Emily Jiang (IBM)
  • Buhake Sindi (Individual)
  • John Clingan (Red Hat)
  • Phil Chung (Oracle)
  • Ed Burns (MSFT)
  • Reza Rahman (MSFT)
  • Arjav Desai (Oracle)
  • Emmanuel Hugonnet (Red Hat)
  • Majid Mostafavi (IJUG)hei

Minutes:

  • Discuss the current status
  • Some integrations with CDI, MP Config, Telemetry were done; RAG is in progress
  • Should we move this to LangChain4j or MP?
  • Reza: Microsoft would like to commit resources to create a CDI extension within LangChain4j itself that mirrors the support for Spring and Java SE currently baked into LangChain4j. This work would have formal backing from Microsoft. However, Microsoft is seeking to do this work in a collaborative fashion with MicroProfile and Jakarta EE vendors. We would like Red Hat, Oracle, IBM, Payara, etc to also commit resourcing and organizational backing to a reasonable extent.
  • John: extension contributions in Langchain4J, what is the MP contribution towards that?
  • Reza: Use MP AI meetings and infrastructure to drive the collaboration for MicroProfile and Jakarta EE vendors.
  • John: MP Telemetry brings in OpenTelemetry and adds extensions. Does Microsoft feel there is a similar need with MicroProfile AI - a spec to integrate LangChain4J into MicroProfile?
  • Reza: The AI space is still very rapidly evolving. This is not yet the right time to standardise anything. The CDI, MicroProfile, and Jakarta EE integrations should be sorted out in LangChain4j, just as it is for Spring. If in the future the areas are stable enough, we can consider standardisation.
  • Reza: Using MicroProfile AI as a place to have conversations, experiments, and staging is appropriate, but the hands-on work should likely happen in LangChain4j, where the infrastructure may not be ideal for MicroProfile and Jakarta EE stakeholders to effectively collaborate.

27th January 2025

Video Recording

Contributors [each attendee must add themself to minutes]

  • Arjav Desai (Oracle)
  • Emmanuel Hugonnet (Red Hat)
  • Buhake Sindi (Individual)
  • John Clingan (Red Hat)
  • Phil Chung (Oracle)

Minutes:

  • OpenTelemetry integrated
  • Fault Tolerance integrated : discussions about separating Langchain4J fault tolerance to be able to use the Microprofile Fault Tolerance retry mechanism
  • New release of LangChain4J that might break compatibility
  • MCP server, does it make sense ?

9th December 2024

Video Recording here

Contributors [each attendee must add themself to minutes]

  • Arjav Desai (Oracle)
  • Emmanuel Hugonnet (Red Hat)
  • Buhake Sindi (Individual)
  • Emily Jiang (IBM)

Minutes:

2nd December 2024

Video Recording here

Contributors [each attendee must add themself to minutes]

  • Arjav Desai (Oracle)
  • Phil Chung (Oracle)
  • Emily Jiang (IBM)
  • Jan Westerkamp (iJUG)

Minutes:

25th November 2024

Video Recording here

Contributors [each attendee must add themself to minutes]

  • Emmanuel Hugonnet (Red Hat)
  • Buhake Sindi (Individual)
  • Arjav Desai (Oracle)
  • Phil Chung (Oracle)

Minutes:

  • Emmanuel mentioned that
  • HTTP transport PR for Langchain4J is still in draft mode and SPI is premature/early for integration.
  • He tried that for Apache HTTPClient as well and had to add more code than Jakarta Rest.
  • Lombok is still being used by some clients on latest 0.36.1
  • Buhake mentioned that
  • Fault Tolerance was already there on the AiService interface but was not triggered. He has made changes for now to make it work with Portable Extension.

18th November 2024

Video Recording here

Contributors [each attendee must add themself to minutes]

  • Emily Jiang (IBM)
  • Emmanuel Hugonnet (Red Hat)
  • Yann Blazart
  • John Clingan (Red Hat)
  • Buhake Sindi (Individual)

Minutes:

  • MP AI will need to use the latest release 0.36
  • Emmanuel used Jakarta Rest with Langchain4J and it worked well locally
  • Buhake: RAG is not working locally for him. CDI injection working with Embedding Model and Embedding Store.

11th November 2024

Video Recording here

Contributors [each attendee must add themself to minutes]

  • Emily Jiang (IBM)

Only Emily was there.

4th November 2024

Video Recording here

Contributors [each attendee must add themself to minutes]

  • Clement Escoffier (Red Hat)
  • Max Andersen (Red Hat)
  • Arjav Desai (Oracle)
  • Emmanuel Hugonnet (Red Hat)
  • Phil Chung (Oracle)
  • John Clingan (Red Hat)

Apologies [each attendee must add themself to minutes]

  • Emily Jiang (IBM) (traveling to W-JAX)

Minutes:

28th October 2024

Video Recording here

Contributors [each attendee must add themself to minutes]

Minutes:

  • Adopt LangChain4J APIs vs new MP AI APIs
  • Buhake: LangChain has many dependencies such as Lombok. If LangChain4J has big changes, MP AI can be protected from these changes. Not sure about the future of LangChain4J. Also integrate with MP Telemetry without impacting LangChain4J. Phase 1: created some facades on top of LangChain4J.

  • Clement: You will have to fork 64 APIs. you will be out of date very fast. What is in the roadmap of LangChain4J v1 (to be released at the end of this year) is the httpclient apis. What about the sdk providing its own clients?

  • Max: work with LangChain4J to detach Lombok from the APIs.  What happens between September and October? Creating new APIs caused more confusion on the choices of APIs. Maybe wait for a while to standardize something. Quarkus did some prototype with LangChain4J and contributed some functionalities to LangChain4J.

  • Arjav: If we focus on defining the integration with MP Rest, we showcase LangChain4J.
  • Rest apis for LLMs
  • Should we influence the LLMs to use the same pattern?

  • For the record: @Tool, @V, @P, @ToolsMemoryId are getting updated in the 1.0 release of Langchain4J (end of 2024)

14th October 2024

Video Recording here

Contributors [each attendee must add themself to minutes]

  • Emily Jiang (IBM)
  • Phil Chung (Oracle)
  • Buhake Sindi (Individual)
  • Yann Blazart (individual)        
  • Emmanuel Hugonnet (Red Hat)
  • Emerson Castaneda
  • Arjav Desai (Oracle)
  • Antoine Sabot-Durand (individual)

Minutes:

Yann and Buhake gave us an update regarding their work about extraction of the API. There seems to be a misunderstanding about the MP AI goal.

Langchain4j brings a lot of limitations : JDK8, lots of bad dependencies. It was often said in this meeting that it shouldn't be a final goal for an implementation. Obviously this information was not shared or understood.

The idea followed by implementers : Yann, Emmanuel, Buhake was to start with Langchain4j to have a first result but in a second time start working on another solution. Taking inspiration from Spring AI for instance…

 

7th October 2024

Video Recording here

Contributors [each attendee must add themself to minutes]

  • Emily Jiang (IBM)
  • Emmanuel Hugonnet (Red Hat)
  • Arjav Desai (Oracle)
  • Phil Chung (Oracle)
  • John Clingan (Red Hat)
  • Buhake Sindi (Individual)
  • Yann Blazart (individual)

Minutes:

  • Next step: since we are happy with LangChain4J, we won’t invest in creating new alternative APIs to LangChain4J. We will be focusing on integrating with LangChain4J.
  • Provide generic SPIs to work with different app runtimes
  • Provide integration with MP Rest Client etc to plug in other Jakarta EE and MicroProfile technologies.
  • Provide facade APIs to work with LangChain4J:
  • Help with LangChain4J to improve the metrics etc.
  • The next piece will be the RAG integration with LangChain4J. Issue here
  • Embedding support
  • In-memory vector store

30th September 2024

Video Recording here

Contributors [each attendee must add themself to minutes]

  • Emmanuel Hugonnet (Red Hat)
  • Arjav Desai (Oracle)
  • Buhake Sindi (Individual)
  • Antonio Goncalves (Individual)
  • Phil Chung (Oracle)
  • Jan Westerkamp (iJUG)
  • John Clingan (Red Hat)
  • Max Rydahl Andersen (Red Hat)
  • Clement Escoffier (Red Hat)

Minutes:

  • All examples were merged to the main branch.
  • Implementation for the current phase are complete.
  • Should we consider release as the first version of alpha release?
  • Overall agreement that reimplementing a new general GenAI library would be wasted effort - working in langchain4j to have SPI’s to enable isolation from vendor specific clients would be good for the broader java system and in smallrye-llm we could focus on extracting a higher level genAI spec (i.e. CDI based beans)

  • About lombok
  • Does not prevent native compilation
  • It’s something that langchain4j wants to remove in the near future
  • HTTP SPI in langchain4j
  • It’s something that should be implemented in Langchain4j to allow plug in different HTTP client, including the MP Rest client

23rd September 2024

Video Recording here

Contributors [each attendee must add themself to minutes]

  • Emily Jiang (IBM)
  • Emmanuel Hugonnet (Red Hat)
  • Arjav Desai (Oracle)
  • Buhake Sindi (Individual)
  • Phil Chung (Oracle)
  • Jan Westerkamp (iJUG)
  • John Clingan (Red Hat)

Minutes:

  • Native compilation
  • LangChain4J uses Lombok for annotation processing. For native compilation, a maven plugin is needed for build-time compilation.
  • There might be some special process needed for this. Not sure…
  • JDK 23 has the annotation processing disabled by default. Have to enable the lombok specially for the later Java.
  • Getting samples to work on Open Liberty
  • Buhake is working on the sample. At the moment, he is debugging a classloading issue.
  • [Jan] Links to AI related Apache Camel blog posts:

16th September 2024

Video Recording here

Contributors [each attendee must add themself to minutes]

  • Emily Jiang (IBM)
  • Yann Blazart
  • Emmanuel Hugonnet (Red Hat)
  • Don Bourne (IBM)
  • Arjav Desai (Oracle)
  • Emerson Castaneda
  • John Clingan (Red Hat)
  • Buhake Sindi (Individual)
  • Phil Chung (Oracle)

Minutes:

  1. Yann, Antoine and Emmanuel are working together on the new proposal on the new AI api/spi. Hopefully in the next 2 or 3 weeks, the draft will be available.
  1. We began with Langchain4J. Due to some technical issues on Langchain4J (on JDK 8, dependency models), it might be a good idea to revisit OpenAI rest api and Spring AI.
  2. Only target BCE for Quarkus, the other app servers can use portal extensions to provide the integration.
  3. Langchain4j is not good with native compilation.
  4. Yann to find out whether Spring AI copes with native compilation.[a]

  1. Yann and Emmanuel work on a defect on the CDI extension
  2. Emmanuel added the support from Wildfly
  3. Buhake to bring the Open Liberty support from a branch to the main
  1. Emily to work with Buhake on the issues Buhake is facing

9th September 2024

Video Recording here

Contributors [each attendee must add themself to minutes]

  • Arjav Desai (Oracle)
  • Emily Jiang (IBM)
  • Buhake Sindi (Individual)
  • Emmanuel Hugonnet (Red Hat)
  • Antoine Sabot-Durand

Minutes:

  1. Discussed the plan of creating some new APIs
  2. Buhake to bring the Open Liberty support from a branch to the main
  3. Emmanuel to add the support from Wildfly
  4. Discussed the new step for the initiative - produce a creation plan for MP AI

19th August 2024

Video Recording here

Contributors [each attendee must add themself to minutes]

  • Emmanuel Hugonnet (Red Hat)
  • Phil Chung (Oracle)
  • Arjav Desai (Oracle)
  • Emily Jiang (IBM)
  • Yann Blazart (individual)
  • Majid Mostafavi
  • Buhake Sindi (Individual)

Current work in

Minutes:

  1. Buhake committed some code changes to smallrye-llm on BCE (https://github.com/sinditech/sindi-langchain4j). He tested on Open Liberty.
  2. Yann - BCE work on Quarkus
  3. Buhake to test his changes on Wildfly and/or Payara. He then propose a PR
  4. Yann to think about creating Integration tests.

12th August 2024

Video Recording (not available)

Contributors [each attendee must add themself to minutes]

  • Buhake Sindi (Individual)
  • John Clingan (Red Hat)
  • Phil Chung (Oracle)
  • Arjav Desai (Oracle)
  • Emmanuel Hugonnet (Red Hat)

Minutes:

Buhake asked about an issue with Open Liberty and CDI with the current Extension. John suggested raising this as a query in either smallrye github discussion or microprofile google groups.

5th August 2024

Video Recording (not available)

Contributors [each attendee must add themself to minutes]

  • Buhake Sindi (Individual)
  • John Clingan (Red Hat)
  • Phil Chung (Oracle)
  • Arjav Desai (Oracle)

Minutes:

We closed out the call in 5 mins as there weren’t enough participants.

29th July 2024

Video Recording

Contributors [each attendee must add themself to minutes]

  • Yann Blazart
  • Buhake Sindi (Individual)
  • John Clingan (Red Hat)
  • Phil Chung (Oracle)
  • Majid Mostafavi (IranianJUG)
  • Arjav Desai (Oracle)

Minutes:

22th July 2024

Video Recording (not available)

Contributors [each attendee must add themself to minutes]

  • Yann Blazart
  • Buhake Sindi (Individual)
  • Emmanuel Hugonnet (Red Hat)
  • Phil Chung (Oracle)
  • Majid Mostafavi (IranianJUG)
  • Arjav Desai (Oracle)

Minutes:

Yann had a conversation wth Emmanuel with the possibility of integrating Wildfly config with Microprofile config. Portability of Wildfly, Quarkus and Jakarta EE profile.

Yann proposed to take the direction to take care of Jakarta EE , microprofile and Quarkus Sides. We can have a core module, then a build compatible extension, a portable one, and a SPI to provide config retrieving capabilities when some servers are not able to use mp-config.

It would be also nice to add some examples with Open Liberty and WildFly in the example module.

15th July 2024

Video Recording

Contributors [each attendee must add themself to minutes]

  • Yann Blazart
  • Buhake Sindi (Individual)
  • Emmanuel Hugonnet (Red Hat)
  • Phil Chung (Oracle)
  • Antoine Sabot-Durand
  • Arjav Desai (Oracle)

Minutes:

Yann start the call be describing his PR on SmallRye LLM : https://github.com/smallrye/smallrye-llm/pull/8

Discussion about bean resolution in a dynamic lookup. We can certainly find another solution here (qualifier, list of bean types etc…)

The extension was created with the Build Time Compatible Extension to make sure that the code is working on Quarkus

8th July 2024

Video recording.

Contributors [each attendee must add themself to minutes]

  • Emily jiang (IBM)
  • Buhake Sindi (Individual)
  • Emmanuel Hugonnet (Red Hat)
  • Phil Chung (Oracle)
  • Majid Mostafavi (IranianJUG)

Minutes:

  • LangChain4J
  • Emmanuel did something similar against Wildfly. It can inject embeddings, LLM and vectorstore. Producer for LLM, vector database, embedding.
  • Global producers? One producer per type; builder pattern
  • Put a PR on smallrye-llm to demonstrate the above idea
  • Embedding APIs? RAG pattern
  • Use CDI extension to create a producer for RAG integration

  • Do we need to create our own APIs or wrap LangChain4J

1st July 2024

Video recording.

Contributors [each attendee must add themself to minutes]

  • Emily jiang (IBM)
  • Arjav Desai (Oracle)
  • Phil Chung (Oracle)
  • Buhake Sindi (Individual)
  • John Clingan (Red Hat)
  • out (Individual)
  • Emmanuel Hugonnet (Red Hat)

Minutes:

  • LangChain4J
  • Antonio talked with LangChain4J community and they are happy for us to add a repo under their org.
  • We need to investigate further on the PoC regarding which CDI beans to produce.
  • We can either keep the code on smallrye-llm or contribute to LangChain4J
  • Produce more code to plug in more LLMs, more examples, more docs.
  • Antonio plans to add more LLMs integration.
  • Emmanuel did something similar against Wildfly. It can inject embeddings, LLM and vectorstore. Producer for LLM, vector database, embedding.
  • Global producers? One producer per type
  • Embedding APIs? RAG pattern
  • Use CDI extension to create a producer for RAG integration
  •  
  • Do we need to create our own APIs or wrap LangChain4J

24th June 2024

Video recording.

Contributors [each attendee must add themself to minutes]

  • Emily jiang (IBM)
  • Arjav Desai (Oracle)
  • Jan Westerkamp (iJUG)
  • Emmanuel Hugonnet (AlpesJUG/Red Hat)
  • Phil Chung (Oracle)
  • Buhake Sindi (Individual)

Minutes:

  • Embedding APIs? RAG pattern
  • Use CDI extension to create a producer for RAG integration
  •  
  • Do we need to create our own APIs or wrap LangChain4J

10th June 2024

Video recording.

Contributors [each attendee must add themself to minutes]

  • Emily jiang (IBM)
  • Antonio Goncalves (Individual)
  • Buhake Sindi (Individual)
  • Phil Chung (Oracle)
  • John Clingan (Red Hat)
  • Arjav Desai (Oracle)
  • JB Roux (Worldline)
  • Emerson Castaneda

Minutes:

3rd June 2024

Video recording.

Contributors [each attendee must add themself to minutes]

  • Emily jiang (IBM)
  • Antonio Goncalves (Individual)
  • Buhake Sindi (Individual)
  • Phil Chung (Oracle)
  • John Clingan (Red Hat)
  • Arjav Desai (Oracle)
  • JB Roux (Worldline)

Minutes:

27th May 2024

Video recording.

Contributors [each attendee must add themself to minutes]

  • Jan Westerkamp (iJUG)
  • Antonio Goncalves (Individual)
  • JB Roux (Worldline)
  • Buhake Sindi (Individual)
  • Emerson Castaneda

Minutes:

  • [Yann] is showing an implementation example

13th May 2024

Video recording.

Contributors [each attendee must add themself to minutes]

  • Antoine Sabot-Durand
  • Clement Escoffier (Red Hat)
  • Arjav Desai (Oracle)
  • Emily Jiang (IBM)
  • Phil Chung (Oracle)
  • Jan Westerkamp (iJUG)
  • Antonio Goncalves (Individual)
  • Jean-Francois JAMES (Worldline)

Minutes:

Emily propose a SmallRye LLM project. Clement agreed. Antoine Said that the scope will be to deliver a framework (maybe a LangChain4j wrapper, maybe a REST client…. still not defined yet) this year and think about as specification from this framework in 2025

  • Repository: https://github.com/smallrye/smallrye-llm
  • GitHub contributors: agoncal, antoinesd, jefrajames,ricken07, yblazart  
    (italic means invitation sent)
  • OpenAI4j is a REST Client for OpenAI (it has been developed by the LangChain4j team as an external library https://github.com/ai-for-java/openai4j)
  • LangChain4j has an SPI to connect to OpenAI, so it’s easy to replace OpenAI4j with other things
  • What do we implement?
  • Chat Model ? Language Model ? Image Model ? Moderation Model ? Embedding Model ?
  • Chat / Image / Video
  • Have the framework deal with the image / video metadata (there’s not much in Java representing an image with Metadata)
  • We could start by generating the Java code from the OpenAI OpenAPI contract so it uses REST Client from MP

06th May 2024

Video recording here.

Contributors [each attendee must add themself to minutes]

  • Antoine Sabot-Durand (individual)
  • Jan Westerkamp (iJUG)
  • Phil Chung (Oracle)
  • Arjav Desai (Oracle)
  • Don Bourne (IBM)
  • Buhake Sindi (Individual)
  • Clement Escoffier (Red Hat)
  • Scott Stark (Red Hat)
  • Ricken Bazolo

Minutes:

Clément explained that most LLM providers are roughly align Open AI API.

A starting point for our vendor neutral MP AI framework could be the implementation of a Rest client able to consume LLM in an abstract approach (Open AI API is a common language at 80% level)

LLM are moving too fast to go for a specification right now. We should wait for 1 year or so to go for it.

  • The goal is to find a common model to consume mot LLM around (see hugging face to have an idea of the numbers of available LLMS) : https://huggingface.co/models?other=function+calling,text-generation-inference&sort=trending
  • Jan and Clement agreed on thinking about a dialect concept (like in JPA) to support this diversity
  • Ricken proposed  to build an abstraction above Open AI. A kind of driver
  • Decision was made to start with a RestClient to consume Open AI
  • Jan suggested that the name of the initiative should be renamed LLM, Clement agreed

29th April 2024

Video recording here.

Contributors [each attendee must add themself to minutes]

  • Emily Jiang (IBM)
  • Antonio Goncalves (Individual)
  • Phil Chung (Oracle)
  • Antoine Sabot-Durand(individual)
  • John Clingan (Red Hat)
  • Clement Escoffier (Red Hat)
  • Yann Blazart(Individual)
  • Arjav Desai (Oracle)
  • Buhake Sindi (Individual)
  • JB Roux (Worldline)
  • Ricken BAZOLO (Individual)

Minutes:

Clément explains the way LangChain4j Quarkus extension is working:

In November 2023 they started building a wrapper around LC4j. But a wrapper wasn’t enough (security propagation, tracing, metrics, etc.). Had 6/7 backward breaks due to LC4J movement. There are more and more users for LC4J. How to make things better for the users, dev mode, etc? Passing emoji or videos was problematic (because streaming does not work well with reactive frameworks like Quarkus). They have difficulties in integrating the Google model (Gemini, VertexAI)

  • Quarkus LangChain4J challenges
  • Integration into its reactive network stack
  • Antonio: How different (simpler?) would it be if there weren't a reactive stack?
    Clement: Not much different. The (Quarkus) API is still blocking. It's not really the reactive stack per se, it's everything else that comes into play (ecosystem integration)
  • Google's model - Integration with rest of stack (metrics for example)
  • Rapid change of LangChain4J
  • Expect this space to be completely different a year from now
  • Rewrite the prompts due to the LLM changes
  • Antonio: The Quarkus LangChain4J wrapper is more complex than expected (thought it would be a few classes, but it is "hundreds"). Clement noted that it's not really that complex. Need to wrap models and vector stores, etc.

22nd April 2024

Video recording here.

Contributors [each attendee must add themself to minutes]

  • Antoine Sabot-Durand (individual)
  • Antonio Goncalves (Individual)
  • Emily Jiang (IBM)
  • Luís Neto (Payara)
  • Yann Blazart
  • Pierre Cesmat
  • John Clingan (Red Hat)
  • Buhake Sindi (Individual)
  • Jean-François James (Worldline)
  • JB Roux (Worldline)
  • Majid Mostafavi(IJUG)

Minutes:

  • Devoxx France experience:
  • Start a impl framework similar to Quarkus Langchain4J extension
  • Option 1: Ask Quarkus to get the langchain4J integrate and make it vendor neutral, e.g. making it to smallrye-ai
  • John to follow up
  • Option 2: Develop our own LangChain4j integration with MP
  • Option 3: Develop our own implementation (but will end up being very similar to LangChain4j)

Option 1

15th April 2024

Video: here

Contributors [each attendee must add themself to minutes]

  • Emily Jiang (IBM)
  • Luís Neto (Payara)
  • Antonio Goncalves (Individual)
  • Antoine Sabot-Durand
  • Don Bourne (IBM)
  • Marie-Alice Blete
  • Emerson Castaneda
  • Phil Chung (Oracle)
  • JB Roux (Worldline)
  • JF James (Worldline)
  • Buhake Sindi

Minutes:

  • Create annotation APIs together with specs - Anotine and Jean-Francios, Ricken
  • TCKs - Buhake
  • LangChain4J is related to LangChain (the Python project) so some terminology/architecture might not be totally “Java compliant” (eg. LangChain4J uses the term “AIMessage” while most of the AI tools would use the term “AssistantMessage” but “AIMessage” is used in the Python version)
  • Do we want to add our own MP API on top of LangChain4J, or just expose the LangChain4J interface?
  • LangChain4J isn’t the cleanest API, but it may be early in maturity terms to build one API that interfaces consistently with all LLMs
  • Have a look at the Quarkus approach to see if there are things we’d want to do to improve

8 April 2024

Contributors [each attendee must add themself to minutes]

  • Antoine Sabot-Durand (SCIAM)
  • Antonio Goncalves (Individual)
  • Don Bourne (IBM)
  • JF James (Worldline)

Minutes:

We started by sharing the existing doc and discussion.

Doc opened 2 weeks ago had a lot of contribution:

https://docs.google.com/document/d/1Nnj3LHJPEMNF_nQDXnyAO_A-0MMOZ73CA9ONHAA5DCQ/edit?usp=sharing

Discussion about the target for V1:

  • Antonio mention that RAG is moving too fast for V1
  • What about having something very basic like EasyRag for a V1
  • MP has a core specification and then standalone specs. MicroProfile AI should be in the Standalone space. They can be backward uncompatible
  • We agreed on the difficulty of TCK
  • Consider use of rouge scores for TCK for RAG implementations

25th March 2024

Meeting was canceled because only 2 people showed.

But Ricken and Antoine started to work on draft regarding what the goal of the spec could be.

The document is here : MicroProfile AI Working Doc

To summarize, our main idea is to have an API that abstract LLM consuming and that would help building RAG.

More to discuss next week.

18th March. 2024

Video: here

Contributors [each attendee must add themself to minutes]

  • Emily Jiang (IBM)
  • Ivar Grimstad (Eclipse Foundation)
  • Emerson Castaneda
  • Antoine Sabot-Durand (SCIAM)
  • Ricken Bazolo (SCIAM)
  • Buhake Sindi

Agenda:

Context: more info here

In the recent MicroProfile Live Hangout (minutes here), Reza brought up the idea of embracing OpenAI libraries in MicroProfile and the idea was welcomed and discussed briefly. At the call, we decided to have some offline discussion on this mailinglist to accommodate the ones who could not make the calls and then further discuss at next week's technical call.

As you may know, recently, Spring has provided Spring AI to create AI applications (see code here) and it seems that a lot of attention was drawn. The other project langchain4j is also very popular, which is to simplify the integration of AI/LLM capabilities. Spring has also provided an example of working with Langchain4j.

I have a few ideas:

1. Provide MicroProfile AI programming model to develop AI applications

2. Provide some APIs if needed to connect with LLMs, which might use Langchain4j

3. Integrate with some other popular OpenAI libraries Context

  • Spring has more interactions with LLMs.
  • Ricken: Mistral IE. Provide a simple way to add architecture. Provide lib. Start transformer; use vector database to bring forward LLMs to Java apps such as Postgres to perform RAG
  • Antoine and Ricken work on the similar projects.
  •  Buhake:
  • Langchain4J - integrated abstract for LLMs
  • Unified model for different LLMs
  • New version of LLMs have changes surfacing up to the app
  • Developers only need to work with interfaces and runtime handles the code changes.
  • Next action:
  • Ricken and Antoine to start a proposal here
  • Buhake to start a proposal
  • Next week, we will bootstrap the proposals

[a]@yann.blazart@gmail.com