Skip to content

bayesimpact/use-cases

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

50 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Bayes Platform — Presentation & Use-Case Library

A shared library of interactive presentations and use-case mockups for the Bayes Platform, Bayes Impact's open-source AI-agent platform for public services. Each page shows the platform applied to a concrete context — a government, a hospital, an employment service — as a ready-to-present deck or interactive mockup. They are illustrations of the product in context, not throwaway demos.

Everything runs on one shared, dependency-free design system: bilingual (FR/EN), mobile-readable, PDF-exportable, published straight to GitHub Pages.

Live → https://bayesimpact.github.io/use-cases/ Reference deck → https://bayesimpact.github.io/use-cases/presentations/bayes/


The one idea that keeps this clean

A new presentation is a declension, not a fork. Every deck is built from the same shared slides, the same design tokens and the same two demo engines. You change content, never the look. That is what stops the library turning into a jungle of clashing styles and conflicting edits:

  • The look, structure and engines live in lib/ (bayes.css, bayes.js, bayes-slides.js). A variant never copies or restyles lib/. Improving the design = one edit in lib/, and every deck inherits it.
  • Each variant is its own folder (presentations/<slug>/), so two people can build in parallel with zero merge conflicts. Decks never touch each other.
  • Shared slides are included by reference (<div data-shared="cover"></div>), with a small JSON block to tweak their wording. You override copy, not CSS.

Follow that and variants stay visually identical to the brand, update together, and never collide. presentations/bayes/ is the reference deck — when in doubt, copy how it does it.


For a non-technical teammate (the easy path)

You don't write code. You describe your use case to Claude Code (opened in this repo) and it builds the deck for you, inside the design system. Give it your raw material and a short instruction:

Prompt to start from (paste your own notes in place of the brackets):

"Create a new Bayes Platform presentation variant for [partner / context, e.g. 'the French Ministry of Health']. Copy presentations/_template to presentations/[slug], follow CLAUDE.md and the reference deck presentations/bayes/, and stay strictly within the Bayes design system and voice. Keep all the shared platform/feature slides as they are; rewrite only the use-case section for my context. Here are my notes, audience and goals: [paste meeting notes, the partner's priorities, any figures or quotes, in FR and/or EN]. Keep copy bilingual FR/EN, mark any invented figure as illustrative, then preview it and show me screenshots."

For the best result, tell Claude: who the audience is, the 2–4 concrete use cases to feature, any real figures, partner names or quotes (and whether they're public), and your language(s). Claude handles layout, styling and the bilingual structure.

Please don't: hand-edit files in lib/, paste in outside CSS or branding, or change colors and fonts. If something looks off-brand, tell Claude — the guardrails are in CLAUDE.md and it will fix it the right way. When you're happy, ask Claude to commit & push; it goes live in ~1 minute.


What stays fixed vs. what you change

You personalize (per variant) Stays shared (never fork)
Cover title/subtitle + the nav section label The design system (lib/bayes.css tokens)
The use-case section — your partner's concrete examples The platform / feature slides (Ancrage, Validation, OS)
The two demo JSON blocks (the conversation & the copilot) The mission, history, model, philosophy, closing slides
Partner logos in assets/, the closing email The two demo engines + all shared slides
Optionally one accent color via the :root hook in <head> Fonts, the pill system, the section-colour language

A variant is essentially the same skeleton with a different use-case middle.


The narrative skeleton (reference deck order)

cover → mission → history → introducing the platform → why it's not a generic agent builder (the 4 bricks) → feature proof (Ancrage · Validation & qualité · platform-OS) → section divider → use cases by vertical (employment · health · government) → our model → team → references → closing

The four bricks the platform is sold on — reuse this language: Suivi des protocoles · Analytique métier · Validation avant production · Ouvert & souverain.


Design system (the non-negotiables)

The look matches bayesimpact.org. All of this is enforced by lib/; don't reinvent it per deck.

  • Tokens (:root in bayes.css): beige --bg #F2EFE9 (Bayes home), ink --ink #000, --cream, gold highlight --gold-bg #EBE092, orange --accent-orange #FF8400, blue --accent-blue #1067FE, green --ok. Fonts Inter + JetBrains Mono.
  • Flat colours only — no gradients. Backgrounds are solid design-system colours.
  • Section-colour language (how the eye reads the chapters): beige = Bayes / platform / model, flat cool --bg-cases = the use-case showcase, flat warm --bg-gov = the government / DPI vertical, flat dark --bg-dark = cover, section divider, closing; deep blue = the mission hero. Returning to beige signals "back to the Bayes story."
  • Pills are outlined with muted text (mirroring the website): a green dot on section-title pills, a grey dot on the cover's flourish chips. Orange is punctuation/accent only — never low-contrast body text; the gold marker (em.hl) is the emphasis highlight.
  • Bilingual: every string is inline FR + EN (<span data-lang="fr">…</span><span data-lang="en">…</span>); the nav language menu toggles them.
  • Voice: airy, concrete, no hype. Mission — "AI to empower people in need, and those who serve them." Demo data is illustrative; flag any invented figure before a real audience.

Build a variant (technical path)

cp -r presentations/_template presentations/<slug>     # slug = kebab-case, e.g. ministere-sante
# then edit presentations/<slug>/index.html only:
#   • <body> data-label-fr / data-label-en  → the nav section label
#   • the cover JSON + the use-case <section>s + the two demo JSON blocks (keep FR+EN)
#   • shared slides stay as <div data-shared="…"></div> (+ optional JSON override)

Preview locally (any static server — the engine is dependency-free):

python3 -m http.server 8016
# → http://localhost:8016/                          splash gate
# → http://localhost:8016/presentations/<slug>/      your deck
  • Shared slides live in lib/bayes-slides.js (cover, earth, platform-*, closing, …). Include with <div data-shared="name"></div>; tweak copy via an inner <script type="application/json"> block. Edit the function once → every deck updates.
  • Two demo engines (lib/bayes.js): a WhatsApp-style conversation (data-demo="chat") and a back-office / verification copilot (data-demo="copilot"), both driven by inline JSON you rewrite.
  • PDF: the browser's Print → Save as PDF (1280×800, backgrounds on) exports a clean deck.
  • Bump the ?v= query on the lib/ includes after changing a shared file, so browsers fetch the new version.

Commit & push → live at https://bayesimpact.github.io/use-cases/presentations/<slug>/.


Publishing & sharing externally

The repo is public and served by GitHub Pages, so any deck is a plain shareable URL — no login, opens on any device, exports to PDF. The site root (index.html) is a splash gate: a visitor enters a short code that routes to /presentations/<slug>/. That's security-by-obscurity only — the repo is public, so never put anything truly confidential in a deck, and mark sensitive pitches as such. To share a deck, send its presentations/<slug>/ URL (or the code for the gate).


Layout

index.html       splash gate (short code → /presentations/<slug>/)
lib/             bayes.css · bayes.js · bayes-slides.js   — the shared system (rarely edited, global impact)
assets/          logos · planet · partner logos · captured incident screenshots
presentations/   _template/  +  one folder per deck (bayes = reference, then each variant)
CLAUDE.md        the full agent playbook (design system, voice, shared slides, engines, checklist)
README.md        this file

Built on the Bayes brand system — Inter + JetBrains Mono; beige #F2EFE9, ink black, orange #FF8400; flat colours, bilingual, open-source, for the public good.

About

Interactive use-case mockups & presentations for the Bayes Platform — Bayes Impact's open-source AI-agent platform for public services

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors