Answers from your documents
RAG chatbot on your own company data
An assistant that looks things up in your manuals, quotes and policies. It is meant to name its source and make nothing up.
Free intro callScope and timeframe depend on your project
An assistant like this looks things up in your own documents before it answers. RAG stands for retrieval‑augmented generation (roughly, “look it up first, then answer”). What matters isn't the AI model, but what happens when the answer isn't in the documents. Our aim is that it then makes nothing up and says it doesn't know. That's less impressive in a demo. In everyday use, it's what counts.
- 01The source is usually shown with the answer, so you can check it
- 02If it finds nothing, it's meant to say “I don't know” instead of guessing
- 03Your access rights can be built in where not everyone should see everything
- 04Connected to SharePoint, Confluence, file shares or databases
- 05Runs in the EU or in your own IT, depending on the project
Connected to
- Claude API
- OpenAI
- LangChain
- Vektordatenbanken
- SharePoint
- Confluence
- PostgreSQL
How it runs
How it runs
A project like this usually runs in four steps.
01
Collect test questions
We collect real questions from your day‑to‑day work. Without them it's hard to tell whether the assistant is any good.
02
Build a prototype on your data
A first version of the assistant, using part of your real documents.
03
Measure against the test questions
Using the test questions, we check how often it's right, how well the sources fit and whether it stays quiet when it has no source.
04
Put it into everyday use
This can include access rights, technical monitoring, cost control, adding new documents and a walkthrough.
What this is not
So you know in advance what it isn't suited for.
- No substitute for well‑kept documents. If the documents contradict each other, so will its answers.
- It doesn't calculate or post anything. It finds and words things, but it doesn't make decisions.
- Not a project for a small set of documents. A good search usually does the job there.
Common questions
Before you ask
We prefer data centres in the EU and providers that don't use your data for training. Depending on the project, AI models can also run entirely within your own IT. What applies to your project is set out in the quote.
That depends on the AI service and its contract terms. We prefer services that don't use your content for training. If you want even more control, AI models can also run within your own IT. Which service suits your project is something we work out together.
That depends mostly on the volume. The AI provider charges for each request it handles, and there's hosting and, where agreed, support on top. Running costs can be estimated in advance from your real volumes. That's best discussed before you start.
That's something to plan for. An AI can be set up so that in uncertain cases it doesn't decide on its own but hands over to a person. Approval before anything is sent or posted, and a log of each step, are possible too. How much control makes sense for your workflow is something we work out together.
Part of
AI integrationAI assistants and automation that read documents, sort enquiries and draft replies, right inside the way you work.
Interested in AI integration?
Drop us a line about what you need. We'll get back to you as soon as we can.
Reach us directlyinfo@tscodeworks.de
We'll get back to you as soon as we can