← Back to work
LIVEAI / Project Intelligence

Cuetoba

A project intelligence platform that organizes project knowledge into reusable, structured context for AI-assisted development.

Role & focus
Product Engineering · Frontend · Backend
Core tech stack
Next.jsTypeScriptTanStack QueryPrismaSupabaseReact Flow
01

The Problem

Project knowledge lives everywhere except where it's needed: docs, tickets, chat threads, meeting notes. When it's time to brief an AI tool or hand context to someone else, it has to be reassembled by hand - every time.

Cuetoba treats project knowledge as structured, reusable material instead of scattered notes - something you organize once and reuse as AI-ready context.

02

What I Built

A live product built as four coordinated applications: a marketing site, a customer workspace for project knowledge and context recipes, a node-based visual flow workspace, and an internal backoffice - all on a shared domain model.

  • Project knowledge workspace: projects, knowledge entries and context recipes that compile project material into structured output - text, markdown, JSON or HTML
  • Visual flow workspace on React Flow: typed nodes (personas, steps, decisions, requirements, documents, questions, systems) edited as a persistent graph, with layout and collision handling
  • Flows as versioned data: snapshot versions with schema versioning, token-based sharing, import/export and portability - not disposable canvas state
  • Review workflow over flow versions: review sessions with items and comments, plus public approve/request-changes links so external reviewers don't need accounts
  • Workspace-level ownership: projects and subscriptions live under workspaces with member roles, not isolated user records
  • Subscription and promotion engine: plans, entitlements, promotions and codes administered through a dedicated backoffice

03Complexity & Concurrency

4 case-specific challenges

Engineering Challenges

Challenge // 01

Structured context, not note storage

Storing notes is easy; making them reusable is the product. Recipes define instructions and an output format, then compile a project's knowledge into context a person or AI tool can actually consume.

Engineering challenge
Challenge // 02

A graph that means something

This isn't generic diagramming. Every node type has domain semantics - a decision is not a requirement is not a question - and statuses like draft, needs_review and confirmed carry product meaning through the flow.

Engineering challenge
Challenge // 03

Versioning a living canvas

A flow is edited continuously, but sharing and review need stable targets. Versions store immutable snapshots with their own schema version, so a shared link or review session stays valid while the canvas keeps moving.

Engineering challenge
Challenge // 04

One product, four surfaces

Marketing, customer app, flow workspace and backoffice share domain concepts - workspaces, projects, plans - without sharing a codebase. The seam between them is the data model, which forced the domain language to stay consistent.

Engineering challenge

04System Blueprint

Service boundaries / data model

Architecture & Contracts

Application surfaces

Client-facing

Marketing

cuetoba.com - positioning & acquisition

Client-facing

Customer App

app.cuetoba.com - projects · knowledge · context recipes

Service boundary

Flow Workspace

node-based project flows

EditorVersioningSharingReview

Backoffice

bo.cuetoba.com - users · workspaces · subscriptions · promotions · audit

Tenant-aware data layer

Data / Persistence

Supabase PostgreSQL

Prisma domain model · Supabase auth

05Engineering Trade-offs

Key Decisions

Deliberate choices made to preserve velocity without sacrificing correctness.

  1. 01

    Recipes as the AI boundary

    Instead of coupling the product to one AI provider, recipes compile project knowledge into portable structured output - text, markdown, JSON or HTML - that works with whatever tool the customer already uses.

  2. 02

    Flows as data, not drawings

    Nodes and edges persist as domain records, and versions store full snapshots with a schema version. Sharing and review operate on versions, so they stay meaningful as the graph evolves.

  3. 03

    Workspace-level ownership

    Projects and subscriptions belong to workspaces with membership roles rather than to individual accounts - the access model matches how a product is actually shared inside a team.

  4. 04

    A real backoffice, not admin routes

    Plans, subscriptions, promotions, users, workspaces and audit live in a separate application, so operational tooling doesn't leak into the customer-facing product.

  5. 05

    Supabase as the product backend

    Auth, PostgreSQL and storage come from Supabase, with Prisma owning the domain model - the effort goes into product surface area instead of infrastructure plumbing.

06Surface Breakdown

Product Surface

Surface / 01CUETOBA

Customer app - app.cuetoba.com

  • Projects & knowledge entries
  • Context recipes & structured output
  • Supabase-backed authentication

Product surface / Cuetoba

Surface / 02CUETOBA

Flow workspace

  • Node/edge graph editor
  • Version snapshots & compare
  • Share tokens & import/export
  • Review sessions & comments

Product surface / Cuetoba

Surface / 03CUETOBA

Backoffice - bo.cuetoba.com

  • Users & workspaces
  • Plans, subscriptions & entitlements
  • Promotions & codes
  • Audit log & dashboard

Product surface / Cuetoba

07Project Status

Current State

Live across three public surfaces - cuetoba.com, app.cuetoba.com and bo.cuetoba.com. The flow workspace is implemented as its own application within the product.

Status: live - shipped and running.

Current project status

LIVE

08Retrospective & Takeaways

What I Learned

  • Context is only reusable when it's structured - freeform notes don't compose.
  • A visual editor earns its complexity when the model underneath is real data, not canvas state.
  • The backoffice is part of the product - operational tooling can't be an afterthought once subscriptions and promotions exist.
  • Multiple focused applications kept each surface honest about its job.

PROJECT INDEX / 09

Related work

All projects →

The next chapter starts here

Let's build something resilient together.