Open Visualization Session Proposal: AI Assisted Dev -- Vibe your Viz

[ryanbaumann]

Proposal

Topic of the session

Empowering the Next 100x Visualization Builders: Making Any AI Assistant a deck.gl Expert

Type of the session

• Collaborate
• Workshop
• Talk

Estimated duration of the session

30 minutes

Date and Time of the session

TBD (October 8 or 9, 2025)

Level

• Beginner
• Intermediate
• Advanced

(Note: This talk covers strategic concepts relevant to all levels, with technical deep dives and demos primarily aimed at intermediate/advanced practitioners and library maintainers.)

Pre-requisite knowledge

Familiarity with the challenges of building complex data visualizations and experience with JavaScript libraries (like deck.gl or similar) is helpful. No prior experience with specific AI coding tools or agent architecture is required, as we will focus on the strategies, practical application, and outcomes.

Describe the session

The barrier to entry for creating complex, GPU-accelerated visualizations with powerful frameworks like deck.gl has traditionally been high. While AI coding assistants (Gemini, Cursor, Claude Code, etc.) are rapidly lowering this barrier and transforming the Developer Experience (DevEx), general-purpose models often struggle with specialized libraries—they hallucinate APIs, miss visualization best practices, or fail to leverage advanced features.

This session explores a practical, DevEx-focused strategy for truly democratizing data visualization: bringing deep vis.gl expertise directly into any AI assistant. We will demonstrate how we can move beyond "vibe coding" and enable builders—developers, analysts, and data scientists alike—to create production-ready, sophisticated visualizations conversationally.

We will dive into the technical strategies and interoperability required to optimize the open-source visualization ecosystem for the age of AI assistance.

Key Topics & Takeaways:

• The DevEx Challenge: Why general-purpose AI struggles with specialized visualization libraries and how this impacts adoption.
• Knowledge Injection Techniques: A practical comparison of RAG (Retrieval Augmented Generation), advanced prompt engineering, and documentation-as-context (e.g., the llms.txt concept) for visualization tasks.
• Live Demos: We will showcase live demonstrations of AI assistants generating complex deck.gl visualizations (using Google Maps Platform as a basemap and data from sources like BigQuery), highlighting how targeted knowledge improves the AI's accuracy and sophistication.
• The Role of Standards: How open standards (like GeoParquet and Apache Arrow) become the crucial interface layer between data sources, AI reasoning, and visualization rendering.
• Architecting for AI: A look at "AI-First API Design"—should we rethink library structures (e.g., favoring declarative over imperative approaches) to be more easily consumable by LLMs?
• AI-Ready OSS Playbook: Concrete takeaways for how the vis.gl community can optimize their libraries, documentation, and examples to accelerate adoption by the next 100x builders leveraging AI.

Session facilitator(s), Github handle(s) and timezone(s)

Ryan Baumann (@ryanbaumann) - PST (Head of Developer Experience, Google Maps Platform)

Meeting notes and Virtual Meeting Link

Meeting Notes: TBD
Virtual Meeting link: TBD

Follow-up / Set-up sessions (if any)

N/A

Additional context (optional)

This session will involve live coding demonstrations and requires standard A/V setup (projector/large display) and a reliable internet connection.

---

[onestardao]

very aligned with this — especially the "can we reason about reasoning?" part.

we’ve been experimenting with visualization pipelines where the AI output isn’t just a surface-level chart or image, but includes:

symbolic decomposition of its own reasoning

visual tracing of failure points (semantic reversal, logic omissions, entropy collapse)

modular hooks for AI logic audit via plugins or post-processing layers

we built a semantic engine (MIT-licensed) to inject this into LLMs and image generation pipelines.
if this overlaps with your goals, happy to share and adapt to your session needs.

[chrisgervang]

Hey Ryan! Thank you for your session submission - sounds like an interesting topic! Just to confirm, are you planning to attend in person? Unfortunately, we won't be able to support remote sessions this year. Looking forward to seeing you at the summit!

[ryanbaumann]

Hi @chrisgervang ! I was planning to, unfortunately I just had a hard personal block for travel now Oct 7-8. I could make it for the 9th, I'll get back to you if that could work.

[chrisgervang]

The 9th works! That's the 2nd day and is the day focused on applications of the frameworks, which fits your session perfectly.

[felixpalmer]

Nice talk, @ryanbaumann! I think there will be some overlap with my session, maybe we sync up a bit nearer the time?

[ryanbaumann]

Thanks Felix, yes happy to sync up. I see some potential overlap but I think the two will work together well - I'd focus on "how to make vis.gl easier to use for agents and new users" while your talk has a specific application of AI Assistance applying vis.gl to your solution. Happy to combine and/or coordinate together!

[chrisgervang]

The GMP MCP assistant looks really useful.. would be very interesting to learn how to make these for all of the OpenVis packages.

[chrisgervang]

Could you please share a headshot to use on our event website?