Building Private, Customizable AI Agents with OSS on Your Own Hardware
Session Description
What if you could harness AI without shipping your sensitive data to external providers or relying on expensive cloud subscriptions? Through this session, I hope to empower developers and builders to build, deploy, and customize powerful AI agents directly on their local machines, leveraging the incredible strength of open-source frameworks.
We'll move beyond abstract concepts to a hands-on, practical exploration of how anyone, from students to seasoned professionals, can create intelligent agents tailored to their specific needs, ensuring data privacy and local control.
Imagine an AI assistant trained on your own documents, a code generator that understands your project's nuances, or a data analyst that keeps your business intelligence strictly in-house – all powered by AI running on your own machine.
Key Takeaways
The Open-Source Advantage in AI: Understand why open-source models like Llama 3.2, 3.3, 4 are not just alternatives, but often superior choices for privacy, customization, and long-term sustainability in AI development.
From Zero to AI Agent: A step-by-step, actionable guide to setting up your local AI environment, including essential tools and platforms from the open-source ecosystem.
Knowledge is Power (Locally): Master techniques to transform your private data – from plain text, Markdown, PDFs, CSVs, to even internal company documents – into powerful knowledge bases for your AI agents through efficient embedding strategies.
Data Sovereignty: Learn how to design and implement AI solutions that keep your data entirely on your machine, ensuring complete privacy and control, a critical aspect for individuals and organizations in India.
References
Session Categories
Which track are you applying for?
Speakers
Harish Kotra
Reviews
Doesn't add a lot of value to the attendees as a workshop, probably not FOSS either - the gaianet-node is proprietary.
I like the AI use case for local use + knowledge search. I don't think I would guide people towards Gaia for this use case because it comes with a bunch of other things that I don't find particularly relevant. e.g. a simple configuration of ollama + a vector db feels like a simpler recommendation for the same use case.
We have concerns about the FOSS nature of the project and the overall value of the workshop.
For future submissions, we recommend that you focus on a project or technology that is fully and transparently open-source. Additionally, you should clearly articulate how your proposal provides unique value beyond what can be achieved with existing, well-known FOSS tools.