An interactive reference architecture for the Databricks Data + AI Platform, specialised for seventy industries on Azure, AWS and GCP, in sixteen languages. Click any box to see what it is, who uses it and where it runs, then export the board as a deck or an editable diagram.
- Seventy industries plus the Standard Reference Architecture, in a searchable picker. Each one specialises the sources, ingestion, teams, consumers, use cases and apps; the platform in the middle stays the same.
- Every board carries ten use cases, four Genie Agents, four AI/BI dashboards and four Databricks Apps written for that industry. Use cases with a Databricks customer story come first.
- Medallion layers in the industry's own terms, linked to its model in the Databricks Industry Data Models.
- Data formats the industry exchanges, such as FHIR, ISO 20022, FIX and DICOM, each linked to its standards body.
- Products: what it is, what the name does not tell you, its capabilities, the docs for your cloud, tours, demos, free Academy courses, certifications, videos and related boxes.
- Use cases: the problem, who benefits, how it is built, the components it uses, customer stories and solution accelerators.
- Teams: the personas inside, the capabilities they use, the use cases they own, and a learning track with certifications.
- Sources: what the system does, who uses it and the data it sends: batch or streaming, its shape, volume and cadence.
- Genie Agents and dashboards: the data an agent reads, its teams and top questions, and the KPIs a dashboard tracks.
- Every chip in a panel links onward, every box is reachable by keyboard, and Escape closes the panel.
- Cloud: Azure, AWS or GCP redraws the cloud services and ingestion, and points every docs link at that cloud.
- Region: pick a region and the board dims what it cannot run yet, and a panel lists every product and feature the docs publish, the same list in every region, with its stage, availability and notes, each linked to that cloud's docs. Sort and filter it like a spreadsheet, and download it as an Excel file.
- Release stage: show GA only, or add Public Preview, Beta, Private Preview and Coming soon. Stages and regions are read from the public Databricks docs.
- Sixteen languages, right to left in Arabic and Hebrew. Exports follow the language.
- Branded or Categorized product names, five platform shapes, thirteen palettes that pass WCAG AA, and light or dark.
- Highlights mark the boxes that carry a customer story, connector, tour, demo, course, certification, video or solution accelerator.
- Live data flows, a platform zoom, a docked details panel, a guided tour and a feedback form.
- Links: every option is in the address bar, so a copied link reopens the same board, and each industry shares with its own preview card.
- Decks: PDF and PowerPoint, with a contents page, the board, and a page for each use case, Genie Agent, dashboard and app. The PowerPoint ends with an editable copy of the board.
- Files: PNG, an animated GIF, standalone HTML, editable draw.io, Visio, SVG and Excalidraw, and a YAML descriptor you can edit and import back.
- Your own copy: Edit clones a board into a tab of your own to rework and export. Tabs and settings are remembered between visits.
- The AI Architecture Assistant builds an editable architecture from a description, a whiteboard photo or a PDF.
- The MCP server gives AI agents all seventy architectures as tools.
Each name opens its board. Add &cloud=aws or &cloud=gcp to open it on that cloud; the
default is Azure.
-
In your Databricks workspace: clone this repository as a Git folder, open
app-installer.ipynband choose Run All. It creates a Databricks App,ideaby default, that needs no catalog, warehouse or data access. -
With the CLI: sync the
appfolder to the workspace, then create and deploy the app:databricks sync app "/Workspace/Users/$USER/idea/app" databricks apps create idea databricks apps deploy idea --source-code-path "/Workspace/Users/$USER/idea/app"
-
Without Databricks:
cd app && python3 main.py, then open http://localhost:8000. The HTML download is a single file that opens anywhere with no server.
A chat sub-app on the same board. Describe a business or use case, or attach a whiteboard photo,
a screenshot, a PDF or a text file. The assistant picks the industry, loads that industry's
template and opens the architecture in an editable tab, leaving the reference board untouched.
app/ai/app.py serves the shared app/index.html, injects ai.css and ai.js, and calls
Claude models on Databricks serving endpoints with the app's own identity, so no API key is
stored. The Fast, Balanced and Thinking tiers map to the newest Haiku, Sonnet and Opus endpoints
in the workspace. See app/ai/README.md.
Gives any MCP client the seventy architectures as four tools: list_industries,
get_architecture, search_architectures and list_resources. It reads the same YAML boards
and JSON resources as the web app. See app/mcp/README.md.
cd app/mcp && python3 -m venv .venv && .venv/bin/pip install -r requirements.txt
claude mcp add arch-explorer -- "$(pwd)/.venv/bin/python" "$(pwd)/server.py"The board is a static folder with no build step: app/index.html renders the industry boards
and reference data it fetches on demand, and writes every export itself.
app-installer.ipynb Databricks App installer, Run All
app/
index.html the board: markup, styles, logic, logos; lazy-loads the rest
arch_schema.js the YAML board-descriptor schema, shared by the app and the tools
industry_icons.js the line-art industry glyphs
main.py static server for Databricks Apps, standard library only
app.yaml Databricks App entry point
architectures/ one YAML board per industry (70) plus manifest.json, fetched on demand
resources/ reference tables as JSON: accelerators, tile aliases, connectors, links,
references, regions, release stages, Academy courses, certifications and
platform doc catalogues, plus each cloud's service icons (resources/logos/)
translations/ one folder per language (English source plus fifteen), fetched on demand
assets/ the Databricks mark, the default share cover, and per-industry cards (assets/og/)
share/ per-industry, per-cloud share stubs (280) that unfurl a preview card
vendor/ js-yaml, the one bundled dependency
ai/ the AI Architecture Assistant: app.py serves the shared board and injects
ai.css and ai.js (see app/ai/README.md)
mcp/ MCP server exposing the reference architectures as tools (see app/mcp/README.md)
docs/ the GIFs in this README, recorded by tools/readme_gifs.js
tools/ generators and gates: palettes (palgen.py), product marks and cloud logos (markgen.py),
the installer (build_installer.py), region data (scrape_regions.py),
release stages and connectors (scrape_stages.py), the Academy,
certification and demo catalogues (build_academy.py,
build_certifications.py, build_demos.py), the platform and content
refreshes (refresh_public_data.py, refresh_content.py with
refresh_videos.py), the outcome-map use-case list (outcome_map.json),
the README GIFs (readme_gifs.js), and the region, stage, height,
layout, link, highlight, label-fit, i18n, export, performance, share
and YAML checks run before a board ships
Read CLAUDE.md first. Every change starts with python3 tools/refresh_public_data.py,
passes the gates listed in CLAUDE.md, and lands as a pull request. node tools/readme_gifs.js
re-records the GIFs above from your working tree; it needs ffmpeg.
Built by Ashraf Osman & Amr Ali. The Apache Spark, MLflow, Delta Lake, Unity Catalog, Apache Iceberg and Delta Sharing marks belong to their respective projects and are used to identify those projects. AWS and the AWS service icons belong to Amazon Web Services; Microsoft Azure, Microsoft Entra, Microsoft Fabric, Power BI, Power Platform and their icons belong to Microsoft; Google Cloud and its product icons belong to Google. Each is used to identify that service, under the provider's terms for architecture diagrams.