Bookcrumb — Product & Systems Architecture

CTX-01

Context

A flat discovery list does not explain why a book fits the reader now. Bookcrumb treats recommendation as an ordered route with memory.

PRB-01

Problem

Mood, theme, pace, profile and history signals had to become one recommendation context that remains consistent across routes, notes and membership state.

ROL-01

My role

Product and systems architecture
Recommendation context
Monetization

SOL-01

Solution

I modeled persistent reader profile and current intent as separate inputs, then built a flow that produces a Reading Route through a Recommendation Context.

BLD-01

What was built

Mood / theme / pace context model
Reading routes and route stops
Reading memory and notes layer
AI-ready provider architecture
Paddle monetization and entitlement flow
PWA and Cloudflare-compatible worker build

ARC-01

System architecture

Recommendation architecture

Reader Profile
Mood / Theme / Tempo
Reading History
Recommendation Context
Reading Route
Books / Notes / Memory

STK-01

Technical stack

Frontend

React 19 · TypeScript · TanStack Start · TanStack Router · TanStack Query · Tailwind CSS 4

Backend / Data

Supabase

AI / Billing

AI SDK · OpenAI-compatible provider layer · Paddle

Runtime

Vite · Nitro · Cloudflare-compatible worker build · PWA

DEC-01

Important engineering decisions

Model recommendation as route context rather than a single-book score
Keep the AI provider layer separate from the product domain
Feed reading history back into future recommendation context

CODE-01

Code / architecture excerpts

Reading route modelTS

01type ReadingRoute = {
02  mood: Mood;
03  pace: ReadingPace;
04  themes: Theme[];
05  stops: BookStop[];
06  memoryContext: ReaderMemory;
07};

OUT-01

Result / current state

Bookcrumb is live with discovery, routes, notes, memory and paid membership combined into one reading experience.

LRN-01

Engineering takeaways

In recommendation products, context design matters more than the model call.
User memory should be an input to the next decision, not just an archive.

OUT-LIVE / Visit product

Bookcrumb

Visit product