AI2026

SiteMind

A RAG-powered AI assistant that retrieves relevant information from websites and user-provided content and turns it into structured answers.

Company: DizaynGrup
Period: 2026 — ongoing
My role: Software developer intern
Focus: Interactive outputs, retrieval, data isolation

Technologies

TypeScript
C#
Python
RAG
Retrieval
LLM Applications

Project overview

SiteMind is a SaaS product that turns websites and user-provided materials into searchable sources for an AI assistant. As part of the development team during my 2026 internship at DizaynGrup, I work on interactive AI-generated interfaces, retrieval quality and project-level data isolation.

From content to a reliable answer

SiteMind has a simple product promise: a user adds a website or their own content, and the assistant grounds its answers in those sources. Behind that straightforward experience is a longer pipeline covering crawling, content selection, processing and retrieval.

Users can choose which discovered pages become part of the knowledge source. Rather than accepting every page by default, the scope of the assistant is controlled at project level.

Showing live progress during longer operations is more than an interface detail; it is part of establishing trust. Letting users see where the system is in the process turns a technical workflow into an understandable product experience.

Content list of crawled pages — The complete content inventory can be reviewed after the crawl.

Page selection — The user decides which pages go into training.

The shape of an answer is part of the answer

The chat surface does not only return plain text. Depending on the question, the model had to produce structured output: a price table when asked about pricing, product cards when asked about products.

A chat DSL in the widget settings defines which response becomes which component. Instead of writing a bespoke interface for every project, the same product foundation can be configured for different needs.

The interactive output layer I contributed to turns model responses into a product experience that is easier to read and act on.

Price table output — A pricing question comes back as a table rather than prose.

Knowing what it doesn’t know

The hardest part of working with RAG was deciding what the model should do when its sources are thin. Making something up is the worst outcome; a flat "I don’t know" loses the visitor.

The product takes a middle path: when evidence is limited, the assistant says so and can hand the conversation to a contact form. It offers a useful next step without hiding the limits of its knowledge.

This work taught me that an AI feature should be judged not only by its correct answers, but also by how transparently and usefully it handles the cases it cannot answer.

Limited evidence and lead form — When evidence is thin the model says so and hands off to a lead form.

The chatbot is only the visible surface

Retrieval may be the product’s distinguishing capability, but the user experience depends just as much on projects, analytics, team management, pricing and authentication flows around it.

Making project identifiers opaque and keeping data isolated by project was one of my less visible but important responsibilities for product trust.

SiteMind showed me that a strong AI capability is not a complete product on its own. Security, data boundaries and reliable everyday flows are what make the model useful in practice.

Widget settings and the chat DSL — The DSL that defines answer shapes — where the product’s scalability comes from.

What this project taught me

This was my first opportunity to apply RAG and retrieval in a product with real user flows. How sources are segmented and how relevant content is selected often affects answer quality more directly than the choice of model.

I also learned that no single metric is enough for evaluation. Reviewing results against real websites and real user questions is an essential part of improving the system.

The project expanded my sense of product responsibility: a component I write must be secure, understandable and useful in a real-world scenario.

Analytics dashboard — A slice of the SaaS layer surrounding the chatbot.