Qualihelp

A company’s own documents, answering its customers, with sources.

Qualihelp is a multi-tenant branded AI SaaS. A company embeds an assistant on its own site with one script tag, uploads its documents, tests retrieval, and publishes answers with citations. We built the platform, the widget and the admin portal.

SaaSAutomation

The Qualihelp tenant dashboard in a browser, with a second window in front showing the assistant on a demo shop’s customer service page

Client: Qualihelp
Sector: AI and customer service
Platforms: Web widget, tenant portal and admin portal
Our role: Full build
Period: Ongoing product partnership

What was the challenge?

A branded AI assistant is only useful if it answers from the company’s own material, cites the source, and does not invent. The platform also has to meter AI cost per tenant, route documents to the right corpus or skill, and let a tenant publish a new version without a developer.

Product: Branded AI assistant
Users: Tenants · their customers · platform support
Type: B2B SaaS

What did bon.do build for Qualihelp?

We built a multi-tenant branded AI SaaS with an ASP.NET Core API, Wolverine, PostgreSQL, S3 and NATS JetStream. The worker extracts text and images, describes them, chunks and embeds documents into a draft corpus version, and the tenant activates a version after testing retrieval in the playground. The widget streams answers over SSE with citations. Every AI call runs inside a per-tenant metering scope.

Ingestion
Text and images extracted, described, chunked and embedded into a draft corpus version.
Retrieval
Hybrid RAG with citations, a retrieval playground and an activation step before anything goes live.
Operations
Per-tenant metering, AI cost, queue health and model control.

Technology

ASP.NET Core
Wolverine
PostgreSQL
S3
NATS
Angular

How did the project run?

01Discover
Answers that hold
What a good answer looks like, and what the assistant must refuse.
02Design
A playground for truth
Retrieval, citations and the activation step designed for a non-technical tenant.
03Build
Ingest, retrieve, meter
A worker, an API and a widget built as one metered system.
04Run
Cost under control
Per-tenant cost, usage and queue health from day one.

Inside the product

The assistant waits in the corner of the customer’s own site. Real product, demo site.
The widget on a phone: every answer shows the documents it came from. Real product, example data.

The Qualihelp assistant embedded on a demo furniture shop’s customer service page, answering with numbered sources — The assistant on a customer’s own site, added with one script tag. Real product, example data.

Qualihelp tenant dashboard with conversations, share of answers from documents, sources per answer and spend — The tenant dashboard: what the assistant answered and what needs attention. Real product, example data.

Qualihelp playground testing retrieval for a question against a draft knowledge version, with the matching passages — Test retrieval on a draft before it goes live. Real product, example data.

Qualihelp knowledge base with versions, their status and document counts — Knowledge versions: activate a new one, or roll back. Real product, example data.

What was the result?

A tenant can upload a document, test retrieval and publish a cited assistant on its own site without a developer. The platform tracks AI cost, usage and queue health per tenant, and operators can control which model each tenant may use.

One platform3surfacesWidget, tenant portal and admin portal.

To embed1tagOne script tag on the customer’s site.

StreamingSSECited answers, streamed as they are generated.