Perplexigrid

The fastest way to build AI-powered analytics in seconds - PerplexiGrid turns natural language into visual dashboards using live web data and the Perplexity Sonar API.

Comment

Inspiration

We wanted to push the boundaries of what’s possible with the Perplexity Sonar API. Everyone’s building chat interfaces, we thought, why not use it for data analytics? What if you could prompt instantly get structured, interactive dashboards instead?

PerplexiGrid was born from the idea that querying the web and building a dashboard shouldn’t be two separate tasks, they should be the same action.

What it does

PerplexiGrid transforms natural language queries into full-fledged, interactive dashboards powered by real-time web search and your own datasets.

You simply say:

“Analyze renewable energy trends in Europe”

And PerplexiGrid builds charts, KPIs, tables, and visual layouts — ready to share, export, or refine. It’s the first analytics tool built entirely on top of and for the Perplexity Sonar API.

How we built it

We designed PerplexiGrid as a full-stack app with: - Frontend: React, Vite, Material UI, Apache ECharts - Backend: Supabase (Postgres + Auth), Express.js - AI Layer: Perplexity Sonar API (Sonsr-Pro model) - Dashboard Engine: Widget system driven by a custom prompt format returning structured JSON

Sonar is the core — we call various prompt-based API modes like f1 (full dashboard gen), l1 (light/embedded mode), r1 (dashboard refinement), and r2 (widget-level tweaks). We tightly control model parameters to ensure output consistency and precision.

Challenges we ran into

  • Structured Output: Getting the model to consistently return structured JSON matching our schema took a lot of prompt tuning.
  • Widget Matching: Mapping user intent to the right visualization was hard, especially with vague or ambiguous prompts.
  • Time Constraints: We couldn’t finish collaborative features in time (real-time editing and chat refinement are in progress).

Accomplishments that we're proud of

  • Stood up the entire app in under a month! We are live!
  • Built the first-ever dashboard platform powered by Sonar API
  • Successfully prompt-engineered a system to generate ECharts-ready JSON schemas from natural language
  • Designed a modular frontend to instantly visualize these schemas
  • Supporting 25+ Visualization types!
  • Created a working product from scratch in just a few weeks, with full prompt to dashboard flow

What we learned

  • Prompting is programming. If you define your schema well and enforce parameters, you can generate complex, structured UIs from plain English.
  • The Sonar API is more powerful than we expected!!! Especially with additional options (i.e. search_recency_filter, and structured response modes)
  • Frontend/AI integration works best when you let the AI speak the language of your components (e.g. our widgets = JSON blocks).

What's next for Perplexigrid

  • Collaboration Features: Multi-user projects, live dashboard editing, comment threads
  • Reference specific Widgets: Refine specific widgets by adding their reference to chat
  • Embed Mode: Add PerplexiGrid to your own apps or Notion pages
  • Styling: Dark mode & custom themes
  • Custom Data Connectors: Google Sheets, Airtable support
  • Marketplace: Share prompt-based templates with others
  • More datatypes: Export and import from JSON, support for PDFs as Datasources

Our long-term goal? Become the Figma for dashboards, powered entirely by AI.

Built With

Share this project:

Updates