Edge and ingress
Azure Front Door with WAF in front of every customer-facing surface; each customer’s custom domain resolved at the edge.
- Azure Front Door
- WAF
- Custom domains
Live in production ·
The AI SaaS we built, run and learn from.
FAQ Hub turns a company’s website and documents into a help centre and an in-app AI assistant. For us it is also a working reference architecture: one platform serving many businesses, answering only from approved sources, safe for AI agents and running on Azure. When we recommend a pattern to a client, it has already met production traffic here.
Architecture
Six layers, one rule: resolve trust before you touch data.
Every request, from a browser, an embedded widget or an AI agent, confirms which customer and product it belongs to from a trusted signal before any content, search or model call is made.
Customer-aware portal
Next.js knowledge portal and embeddable widgets (chat, inline help, Command-K search) that confirm which customer they serve before reading any published content.
- Next.js
- React 19
- Per-customer cache
Application core
ASP.NET Core and a headless CMS on Azure Container Apps. Sign-in, customer and product are confirmed before any business rule runs.
- ASP.NET Core
- Container Apps
- Azure AD B2C
Data and search
Azure SQL and Blob Storage for content and control data; keyword and vector search for grounding.
- Azure SQL
- Blob Storage
- Vector search
AI and business services
Azure AI for generation and judging, Stripe for metered plans, Azure Communication Services for notifications.
- Azure AI
- Stripe
- ACS
Agent surface
An MCP server with product-scoped API keys, least-privilege presets, dry-run plans, audit and explicit publish.
- MCP
- Scoped keys
- Audit ledger
Patterns we reuse for clients
What FAQ Hub taught us, applied to your platform.
Grounded answers with a no-answer path
Answers come only from published content, cite their sources, and hand off to a person with context when evidence runs out.
Agents that cannot bypass review
Authoring agents export, plan and apply drafts through MCP tools. Publishing stays an explicit, auditable step.
Customer data kept strictly separate
Which customer a request belongs to comes from a verified key or domain, never from anything an AI writes.
Open knowledge out
Published knowledge renders as Open Knowledge Format v0.2 bundles: Markdown with typed frontmatter, status and verification.
Metered AI usage
AI conversations are measured per customer and mapped to plan allowances, so pricing tracks the real cost of AI.
Paved-road delivery
Azure Pipelines take every change from commit through tests to Container Apps; the marketing site ships the same way to Cloudflare.
For product teams
Need AI support in your own product?
FAQ Hub is available as a product. Start a free trial on faqhub.io, or talk to us about an enterprise rollout with custom integrations.
Want this architecture under your product?
We bring the same data separation, trusted answers and AI safety patterns to your platform, starting with an AI readiness assessment.