Live product · Independent end-to-end delivery
My Visual Path
Making everyday steps easier to see.
I designed, built, and deployed a free visual-support website where families can find a routine or social story, adapt it, and print it or use it on screen. The project brings together my product thinking, frontend engineering, API design, data modeling, content workflow, and production delivery.
- My role
- Product, design & full-stack engineering
- Released
- October 2026
- Core stack
- React · Node.js · PostgreSQL · Cloudflare
01 / Built around everyday use
A practical tool, shaped by a parent’s perspective
I wanted families and supporting adults to have a straightforward way to prepare for everyday routines and unfamiliar situations. I translated that need into reusable visual schedules and social stories, with a short path from finding a resource to using it.
Start with something useful
Ready-made templates, categories, search, related routines, and recently viewed resources help people find a starting point.
Adapt to the situation
Builders let people change wording, choose visuals, and reorder activities or story pages rather than rebuild everything from scratch.
Use it in daily life
Saved routines, progress, a story reader, and printable PDFs support use on a device or away from a screen.
Keep the feedback loop open
Anonymous helpfulness feedback and resource requests give families a way to suggest what is missing without an email signup.
Clarity, accessibility, and privacy as design decisions
I chose simple pictograms, minimal backgrounds, literal actions, neutral expressions, and consistent characters to keep attention on the step being explained. Editable text, predictable sequences, guided tours, and several print sizes give supporting adults ways to adapt resources to their context.
The core resources are free and public browsing does not require an account. Personal routines and stories are saved in browser storage, while the server holds the shared catalog and community submissions. I avoided requiring children’s personal information. These choices describe the product’s approach; they are not a claim of accessibility certification or clinical effectiveness.
02 / What I delivered
Ownership across the whole product
Product & content
Defined the everyday-use journeys, organized reusable activities and story visuals, and created and reviewed templates and illustrations.
Frontend engineering
Built discovery, customization, local saves, interactive use, sharing, feedback, guided tours, and printable PDF layouts in React.
Backend & data
Designed the Express API, relational catalog, validation, protected administration, image pipeline, and transactional CSV imports.
Delivery & operations
Connected Cloudflare hosting, DNS and R2, a Render API, and Neon PostgreSQL; configured environments, CORS, HTTPS, sitemap delivery, and SEO prerendering.
The result is a deployed product that connects a real user journey to the systems behind it: browse a shared template, customize a personal version, save it locally, and generate a document for daily use.
03 / Engineering decisions
Clear boundaries between UI, data, and files
| Layer | Responsibility |
|---|---|
| React in the browser | Discovery, editing, interactive use, local saves, and PDF generation. |
| Node.js / Express API | Catalog delivery, submissions, authorization, administration, and image processing. |
| Neon PostgreSQL | Shared activities, templates, ordered relationships, story pages, categories, feedback, and import records. |
| Cloudflare R2 | Image files referenced by catalog URLs; images are separate from the JSON records. |
| Browser storage | Personal routines, compact stories, preferences, and progress for the same browser origin. |
React architecture
React + Vite + Router
Feature pages for routines and stories, shared UI and hooks, client-side navigation, and an application error boundary.
React Query + Axios
Server catalog fetching, keyed caching, loading and error states. Default stale time is five minutes, with one retry and no focus refetch; Axios uses a 45-second timeout.
Redux Toolkit + localStorage
Saved routines, compact story records, and schedule progress. Unsaved edits stay in component state; personal saves remain in the current browser.
dnd-kit + driver.js
Reordering activities and story pages, plus guided builder tours that explain the editing flow.
React PDF + Buffer
Dynamically loaded document generation in the browser, with image preparation, duplicate-export protection, download, and print preview.
Helmet + prerendering
Page metadata, canonical URLs, structured data, and optional Playwright-based HTML capture for public pages and template slugs.
Story storage keeps page references and overrides instead of repeating catalog image metadata. Saved records are resolved against API/cache data. Local saves do not provide cross-device synchronization or guarantee full offline use; clearing browser storage can remove them.
Node.js architecture
Express 5 + PostgreSQL
Public catalog reads, community submissions, and protected admin routes. pg connection pooling and parameterized queries keep database access separate from HTTP handling.
Clerk authorization
Admin endpoints require a Bearer token and an admin role. The public React app has no Clerk dependency or sign-in requirement for browsing.
Validation & safeguards
Zod schemas validate catalog inputs, identifiers, filters and pagination; some submission routes use manual validation. Helmet, configured CORS, 64 KB JSON limits, and route-specific rate limits add safeguards.
Multer + Sharp + R2
Accept an image up to 5 MB, decode and validate it, limit pixel count, correct orientation, and resize within 1600 × 1600. The S3-compatible SDK writes to Cloudflare R2.
Content integrity
Conditional object writes and SHA-256 metadata reuse matching images and avoid overwrites. CSV preview and transactional imports track import keys and content hashes.
Operations & diagnostics
Request IDs, status and timing logs, generic server errors, query timeouts, and graceful shutdown. OpenAPI documentation is exposed in development; PGlite supports database-backed tests in the source.
General traffic is limited to 300 requests per minute, admin traffic to 60, resource requests and general feedback to five each, and template feedback to 30. Counters are process-local. CORS controls browser-origin access; authentication and role checks protect administration.
Relational content model
Routine templates connect to activities through an ordered join table. Social-story templates contain ordered pages with text and optional visual references. Categories organize both catalogs. Separate submission and feedback records support administration, while slug aliases preserve older content URLs.
Three journeys through the stack
- Browse a template. React Query checks its cache; Axios requests the public API when needed; Express queries PostgreSQL and returns JSON; the browser loads images from their public URLs.
- Save a personal version. The builder validates edits, writes a browser record, updates Redux, and makes the saved routine or story available in the personal workspace. This does not modify the shared catalog.
- Print or download. The export hook loads PDF code, fetches and prepares images, renders a PDF Blob, and opens a download or print preview. Image CORS must allow this processing; merely displaying an image is not enough.
Explore the full source-based architecture guides
The companion guides include package responsibilities, route maps, storage boundaries, and interactive request sequences. They document inspected source, rather than a live infrastructure audit.
04 / From source to production
Delivery, discoverability, and maintainability
I connected the frontend to Cloudflare hosting, the Express service to Render, PostgreSQL to Neon, and images to R2. Environment configuration separates public frontend settings from server secrets. The public site and protected admin workflow use different access boundaries.
SEO beyond metadata
Canonical URLs, public template slugs, structured data, a sitemap, and optional prerendered HTML make catalog content available to crawlers. Personal builder pages use indexing boundaries rather than compete with public templates.
Cloudflare delivery
Vite builds the frontend assets. The supplied Worker serves static assets and forwards the sitemap to the backend with a one-hour cache header. Browser API calls go directly to the configured API origin.
Maintainable content operations
Protected catalog management, image processing, CSV previews, transactional imports, and reporting keep shared content maintainable after launch.
Production tradeoffs
The 45-second client timeout accommodates the observed hosting cold start; it does not eliminate the delay. Browser saves reduce account friction but require clear storage expectations.
I submitted the production sitemap and checked public structured data during launch. The architecture guides also identify test and validation tooling, but do not establish that every test passed or that a formal accessibility audit was completed.
05 / Next: AI approach · Planned
Assistance with a reviewable result
My proposed next step is to help a supporting adult create a first draft faster while keeping the existing curated catalog and manual editing flow useful on their own. This is a roadmap, not a feature currently shipped in My Visual Path.
- Start with existing resources. Retrieve relevant activities, story pages, and approved visuals for a requested situation before proposing new content.
- Generate a structured draft. Ask the model for short steps and text in a defined schema, referencing existing resource IDs when possible. Validate the response on the server before it reaches the builder.
- Require adult review. Let the supporting adult edit, reorder, replace, or discard every suggestion before saving, printing, or publishing it.
- Review new imagery separately. For missing visuals, explore controlled prompts with consistent characters, literal actions, and minimal backgrounds. Newly generated assets would need manual review before entering the shared catalog.
- Evaluate before release. Check relevance, plain wording, visual consistency, invalid resource references, privacy, latency, and cost using representative scenarios. Keep provider keys server-side and use request limits and usage budgets.
The intended boundary is educational visual support. The planned assistant would not diagnose, recommend treatment, or need a child’s name or diagnosis. Provider selection, retrieval method, limits, and pricing remain implementation decisions.
A product I can explain from interface to infrastructure
My Visual Path demonstrates how I turn a practical need into an end-to-end web product, make explicit engineering tradeoffs, and carry the work through production delivery.