GSoC '26 - Reaching out to mentors for Project #13: No-Code AI Workflow Automation with n8n and OVMS

[naraen-ram]

Hi @praveenkk123 and @mjdomeik,

I'm Naraen Rammoorthi, a 2nd year Computer Science student. Project 13 caught my attention, it's the kind of project that makes me genuinely excited. I've been contributing to the OpenVINO codebase and I currently have two PRs under review (#3958 in NNCF and #33842). Recently, I've been working on automating workflows using n8n out of personal interest.

My understanding of the project: build a custom n8n community node that connects to OVMS, exposing model selection, device targeting (CPU/GPU/NPU/AUTO), and inference execution directly from the n8n canvas. On top of that, implement a document processing workflow that routes jobs intelligently across devices using the OpenVINO AUTO plugin, with the full stack deployable via Podman Compose on Intel AI PCs.

I've been exploring the integration this week by getting OVMS running locally and made successful inference calls against the KServe v2 REST API. My current approach is direct OVMS integration where the node constructs the KServe v2 payload itself rather than routing through a Python gateway. I'd love your input on whether this aligns with the intended architecture or if a gateway layer is expected as part of the scope.

I plan to have a working prototype ready soon and will share it here once it's at a stage worth showing. Looking forward to hearing your thoughts on the approach.

Thanks!

[naraen-ram]

Hi @praveenkk123 and @mjdomeik,

Just an update on the prototype.
Here is the github repo: https://github.com/naraen-ram/n8n-ovms

The prototype is a working custom n8n community node that connects directly to OpenVINO Model Server via the KServe v2 REST API. To demonstrate it end to end, I built a face detection workflow in n8n — an image is preprocessed into a float tensor outside n8n, passed into the workflow via a Code node, sent to OVMS for inference, and the model detects a face returning structured results with bounding box coordinates as clean JSON.

The current gaps I'm aware of and plan to address to further improve the prototype:

• Input preprocessing helpers so users can pass images or text directly instead of raw float tensors — this is the biggest usability gap right now
• Full AUTO plugin observability where executed_on reflects the device OpenVINO actually routed to at runtime, not just what was requested
• A Podman Compose file to bring up n8n and OVMS together with a single command instead of two separate Docker run commands

The node currently uses direct OVMS integration with no gateway in between, and preprocessing is handled upstream using n8n's own Code nodes. Is this the intended architecture or do you see a case for a dedicated tokenization/preprocessing service?

Thanks!

[naraen-ram]

Hi @praveenkk123 and @mjdomeik,

This is my updated prototype: https://github.com/naraen-ram/n8n-ovms

I've pivoted from a face-detection model to a text-detection model. This aligns much better with the actual project scope of building an end-to-end document processing pipeline.

The current prototype shows a full workflow where a PDF is inputted through an n8n webhook, processed via a lightweight preprocessing service to generate model-ready tensors, and then sent to OpenVINO Model Server for inference using a custom n8n node. The workflow is fully containerized using Docker Compose, so the entire setup can be run with a single command.

I’ve focused on keeping the system modular, with preprocessing, inference, and orchestration handled as separate components, so that additional steps like OCR or LLM-based classification can be added later without major changes to the pipeline.

I would really appreciate your input on this prototype as it would help me in drafting my formal proposal.

Thanks,
Naraen

[praveenkk123]

Hi Narean,
Great start on the project! Please submit a proposal that outlines how you plan to incorporate advanced features such as Agentic AI Nodes, custom code nodes, and any other recent innovations in this area that may benefit this initiative.
At this stage, please do not implement these enhancements. Simply provide your proposal, and once if it is approved, you can begin implementing the selected features when the actual program starts.

Thanks
Praveen

[naraen-ram]

Hi @praveenkk123 and @mjdomeik,

Right now, I’m planning to extend the document processing pipeline to support more adaptive workflows, where outputs from OVMS models can be used to dynamically trigger the next steps (for example, detection -> OCR -> classification), with an LLM or n8n’s AI agent acting as a decision layer to guide the workflow.

I’m also thinking about an alternative approach where a “controller” node decides which nodes to call and in what order at runtime, instead of relying entirely on a predefined workflow.

Does this align with what you have in mind for the project, and is there anything important I might be missing from the expected scope? Also, if possible could you share any OpenVINO proposals from the previous GSoC years, it would give me a better idea on how to structure and write my proposal.

[mjdomeik]

Naraen, Let’s wait until Praveen returns (next week) to get a response on your idea. I’ve forwarded your request for past proposals to our adminstrators to see if we can share. Stay tuned there. Thanks! Max From: naraen-ram ***@***.***> Sent: Wednesday, March 25, 2026 10:40 AM To: openvinotoolkit/openvino ***@***.***> Cc: Domeika, Max J ***@***.***>; Mention ***@***.***> Subject: Re: [openvinotoolkit/openvino] GSoC '26 - Reaching out to mentors for Project #13: No-Code AI Workflow Automation with n8n and OVMS (Discussion #34710) Hi @praveenkk123<https://github.com/praveenkk123> and @mjdomeik<https://github.com/mjdomeik>, Right now, I’m planning to extend the document processing pipeline to support more adaptive workflows, where outputs from OVMS models can be used to dynamically trigger the next steps (for example, detection -> OCR -> classification), with an LLM or n8n’s AI agent acting as a decision layer to guide the workflow. I’m also thinking about an alternative approach where a “controller” node decides which nodes to call and in what order at runtime, instead of relying entirely on a predefined workflow. Does this align with what you have in mind for the project, and is there anything important I might be missing from the expected scope? Also, if possible could you share any OpenVINO proposals from the previous GSoC years, it would give me a better idea on how to structure and write my proposal. — Reply to this email directly, view it on GitHub<#34710?email_source=notifications&email_token=AAKC5ZJX2VIWKYU4K376PXD4SQKXHA5CNFSNUABIM5UWIORPF5TWS5BNNB2WEL2ENFZWG5LTONUW63SDN5WW2ZLOOQXTCNRTGA4TINJZUZZGKYLTN5XKO3LFNZ2GS33OUVSXMZLOOSWGM33PORSXEX3DNRUWG2Y#discussioncomment-16309459>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AAKC5ZI7HCKALRSSEZGPRMT4SQKXHAVCNFSM6AAAAACWSSZ4OWVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTMMZQHE2DKOI>. You are receiving this because you were mentioned.Message ID: ***@***.******@***.***>>

[naraen-ram]

Hey Max,
Thank you for responding. Since the application deadline for GSoC is closing in, I will send my proposal draft by mail to both you and Praveen soon. I would really appreciate your feedback on the draft so I can make necessary changes.
Could you please send your email id?

Thanks,
Naraen

[mjdomeik]

Naraen, We unfortunately cannot share past proposals, however the link here provides what we can share. https://github.com/openvinotoolkit/openvino/wiki/Google-Summer-Of-Code This includes guidelines for the proposal as well as past completed projects. Thanks! Max From: Domeika, Max J Sent: Wednesday, March 25, 2026 5:49 PM To: 'openvinotoolkit/openvino' ***@***.***>; openvinotoolkit/openvino ***@***.***> Cc: Mention ***@***.***> Subject: RE: [openvinotoolkit/openvino] GSoC '26 - Reaching out to mentors for Project #13: No-Code AI Workflow Automation with n8n and OVMS (Discussion #34710) Naraen, Let’s wait until Praveen returns (next week) to get a response on your idea. I’ve forwarded your request for past proposals to our adminstrators to see if we can share. Stay tuned there. Thanks! Max From: naraen-ram ***@***.******@***.***>> Sent: Wednesday, March 25, 2026 10:40 AM To: openvinotoolkit/openvino ***@***.******@***.***>> Cc: Domeika, Max J ***@***.******@***.***>>; Mention ***@***.******@***.***>> Subject: Re: [openvinotoolkit/openvino] GSoC '26 - Reaching out to mentors for Project #13: No-Code AI Workflow Automation with n8n and OVMS (Discussion #34710) Hi @praveenkk123<https://github.com/praveenkk123> and @mjdomeik<https://github.com/mjdomeik>, Right now, I’m planning to extend the document processing pipeline to support more adaptive workflows, where outputs from OVMS models can be used to dynamically trigger the next steps (for example, detection -> OCR -> classification), with an LLM or n8n’s AI agent acting as a decision layer to guide the workflow. I’m also thinking about an alternative approach where a “controller” node decides which nodes to call and in what order at runtime, instead of relying entirely on a predefined workflow. Does this align with what you have in mind for the project, and is there anything important I might be missing from the expected scope? Also, if possible could you share any OpenVINO proposals from the previous GSoC years, it would give me a better idea on how to structure and write my proposal. — Reply to this email directly, view it on GitHub<#34710?email_source=notifications&email_token=AAKC5ZJX2VIWKYU4K376PXD4SQKXHA5CNFSNUABIM5UWIORPF5TWS5BNNB2WEL2ENFZWG5LTONUW63SDN5WW2ZLOOQXTCNRTGA4TINJZUZZGKYLTN5XKO3LFNZ2GS33OUVSXMZLOOSWGM33PORSXEX3DNRUWG2Y#discussioncomment-16309459>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AAKC5ZI7HCKALRSSEZGPRMT4SQKXHAVCNFSM6AAAAACWSSZ4OWVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTMMZQHE2DKOI>. You are receiving this because you were mentioned.Message ID: ***@***.******@***.***>>

[naraen-ram]

Hey Max,

Thanks for getting back to me on this. I have completed my proposal draft and I have gone ahead and sent it to you and Praveen by mail. Please go over it and let me know what you think.

I've sent the proposal draft to [max.j.domeika@intel.com]. If this is not the correct email ID please do let me know.
Please feel free to reach out to me at [naraen13@gmail.com].

Thanks,
Naraen