AI as operator, not widget
Chat is the interface for actions: documents, tasks, messages. The model must understand context and trigger flows — not only generate text.
← Back to project
Case study · Web + Mobile App
An AI product that goes far beyond a chatbot
Harvey was a substantial product build: an AI core, subscriptions, role-based access and a full module system — from canvas and documents to invoices, tickets and chat.
- Project
- Harvey
- Client
- Harvey
- Year
- 2026
- Role
- Product engineering (React Native + Next.js + AI)
AI
Claude + OpenAI
RBAC
Roles & permissions
Subs
Subscription model
7+
Core modules
Problem
Harvey was not “another AI chat”. The goal was a personal AI operator — an app that helps run your day on your behalf: messages, documents, tasks, invoices and support, with the feel of a product rather than a demo.
That kind of project grows complex fast: authentication, subscriptions, different roles, files, conversations, admin and reliable AI behaviour. Without clear architecture you end up with a pile of features that do not connect.
We needed a mobile-first experience with a strong AI core and a backend that can scale — plus modules that are useful alone and together form an operating system for the user.
Approach
We built in modules around one AI core. Claude and OpenAI cover conversation and actions; Supabase stores users, roles and data; Next.js APIs on Vercel orchestrate agents and integrations; React Native (Expo) carries the mobile experience.
Subscriptions and role-based access were part of the design early — not “we will add billing later”. Without them an AI product has no usage boundaries and teams cannot safely share access.
Subscriptions + RBAC from the start
A subscription model bounds and monetises usage; roles separate owner, members and admin. That is the foundation of a real product, not a toy.
Modular product surface
Canvas, documents, invoices, tickets and chat are separate modules with a shared AI and data layer — easier to ship and clearer UX.
Product map
- Auth + Roles
- Subscriptions
- AI Core
- Canvas
- Documents / Files
- Invoices
- Tickets
- Chat
What shipped
AI core
A conversational operator with Claude/OpenAI — context understanding, task execution and links into app modules.
Subscriptions and roles
A subscription flow for feature access and role-based permissions for safe sharing inside a team/account.
Canvas and documents
A workspace plus file/document management — where AI and user collaborate on real material.
Invoices, tickets, chat
Operational modules for invoices, support tickets and communication — so Harvey covers more than a single chat screen.
Results
Harvey grew into a full AI application with real product depth: not just prompt demos, but a system with billing, files, roles and subscriptions.
For the client that means a foundation the product can grow on — with architecture that can take new modules, instead of starting over.
- Substantial scope: AI + billing + RBAC + modules
- Mobile app (Expo) + web/API layer
- Canvas, documents/files, invoices, tickets, chat
- Built for product iteration, not a one-screen MVP
Who this approach is for
This case study shows that Tim Blažič can lead complex AI and product builds — not only marketing sites. If you need a web/mobile app with AI features, subscriptions, roles and multiple modules, this is relevant proof.
A good fit for startup teams and companies that want a solopreneur developer with end-to-end ownership: product decisions, UX, React Native/Next.js, Supabase and AI integrations — with direct communication and a clear scope.
Want a similar project?
Write to me — I'll reply within 24 hours with a free consultation and a clear next step.