GSoC '25: Reaching out to mentors for discussion about the OpenVINO Messenger AI-Assistant Project

[AJThePro99]

Hello everyone!

And hello mentors @riacheruvu & @DimaPastushenkov

I'm Aadith, a sophmore undergraduate from India, currently working on a proposal for the OpenVINO Messenger AI-Assistant for an AI PC for GSoC 2025. I'd like to share my current approach and get your feedback to ensure it aligns with the project's vision.

My pre-requisite PR for this project is #29045 closing the issue #20547. The PR is complete and pending review for merging.

---

From my understanding of the project, it involves:

1. Fetching messages from a popular messenger
2. Storing the messages locally in a secure, efficient way.
3. Implementing a RAG pipeline to enhance a local LLM with the stored chat data. The basic features will include a Daily digest tab, and a Q&A Tab.
4. Optimizing inference using OpenVINO on the AI PC's GPU.
5. Providing a UI (Using the Qt framework) for interaction with the assistant.

As of right now, I'm working on three proof of concepts(PoCs):

• PoC#1: Retrieving the user message data using an API.
• PoC#2: Using the relevant chat data provided by the RAG to generate personalized summaries.
• PoC#3: Integrating the above PoC's to build a working pipeline and deliver a minimum viable product.

After that, the focus will be on following proper development practices to make a scalable, adaptive, modular and potentially, plugin supporting application.

---

Questions for mentors on the project:

1. Since chat data is stored locally, I'm considering encryption methods.

• What encryption methods and access control strategies would you recommend to ensure the security and privacy of user data?

2. To enable easy integration of additional messaging platforms in the future, while also making feature development easy

• Would a modular plugin-based architecture be advisable? If so, what design patterns or practices would you suggest?

3. Given the potential size of chat data and the embeddings,

• What strategies should I use to avoid memory overflow or performance degradation, especially when storing embeddings?

4. Based on your experience, what common challenges should I anticipate during the development of this project (especially the integration of OpenVINO and LLM pipeline), and how can they be mitigated?
5. On what criteria will the success of this project will be evaluated upon? And could you suggest some ways in which I can align my approach accordingly?
6. Is my current approach aligned with the original vision of the project, or would you recommend any adjustments?

---

Questions for mentors regarding GSoC:

1. Could you share your preferred method of communication for sharing as well as reviewing my proposal draft once it's ready?
2. Are there specific formats or elements you’ve found effective in past GSoC proposals that you'd recommend I include?
3. Are there any resources, documentation, or previous GSoC projects that you think would help me refine my approach?

---

I would greatly appreciate any insights, suggestions, or resources you could share. Looking forward to your feedback!

Thank you!

[AJThePro99]

Here is my rough draft for a high level system architecture design for this project. I'm going to improve upon this diagram as time passes, and my understanding of the project becomes more solid.

• The red line represents the "User Query Processing Flow" — how the system behaves when handling user inputs, retrieving context, and generating responses.
• The purple line represents the "Daily Digest Flow" — how the system aggregates and processes messages to create daily summaries.

---

---

I'd appreciate any feedback or suggestions on how to enhance this architecture!