A multi-tenant AI product demonstration — tenancy, roles, metering and billing designed alongside the AI feature rather than after it.
- Client
- Helio
- Industry
- Technology & SaaS
- Year
- 2026
- Services
- AI SaaS DevelopmentSaaS DevelopmentFull-Stack Development
Overview
What this project is.
Helio models a subscription product whose core value is an AI capability. It exists to show the parts of AI SaaS that demos usually skip: how customers are separated, how usage is metered, how the feature is priced, and how quality is measured once real people depend on it.
Challenge
The problem to solve.
AI features have variable, usage-driven running costs. A product priced on seats but delivered on tokens can become less profitable as it succeeds — and by the time that shows up in the numbers, the pricing page is already public and the architecture assumes the wrong model.
Insight
What we noticed.
Inference cost per customer is a product decision, not an infrastructure detail. Deciding what to meter, when to cache, which model handles which request and where to cap context has to happen while the architecture is still soft.
Strategy
The approach taken.
Build the commercial foundations first — tenancy, roles, metering, plans — then the AI capability on top of them, with cost per interaction visible from the first prototype. Route inexpensive requests to smaller models, cache aggressively where answers repeat, and surface usage to the customer rather than hiding it.
Creative Direction
How it looks and moves.
A calm, dense product interface built for people who will use it every day: strong typographic hierarchy, restrained colour reserved for state and action, and data presentation designed as a first-class part of the system rather than left to a chart library default.
Execution
What the work involved.
Multi-tenant data model with row-level isolation
Authentication, organisations and role-based access
Plan management, usage metering and subscription billing
AI feature with retrieval, caching and model routing
Evaluation harness run against every prompt change
Admin tooling so support does not require engineering
Deliverables
- Multi-tenant application
- Billing & metering
- AI feature layer
- Evaluation harness
- Admin console
- Deployment pipeline
Stack
- Next.js
- TypeScript
- FastAPI
- PostgreSQL
- Redis
- OpenAI
- Docker
- AWS
What it sets out to achieve
The outcome the work is built for.
This is a capability demonstration, so the outcomes below describe design intent rather than measured performance. We publish numbers only where a client has verified and approved them.
Priced on what it costs
Metering and cost-per-interaction visibility exist before the pricing page does.
Sellable, not just demonstrable
Accounts, roles, billing and admin tooling are part of version one rather than a later project.
Quality with a number on it
Prompt and retrieval changes are scored against a fixed set instead of judged by impression.
