AI consulting

An AI chatbot that answers with your own knowledge

A general language model does not know your products, price lists or internal rules. With retrieval-augmented generation (RAG), the chatbot first looks up the relevant passages in your own documents and answers on that basis, with a link to the source. Sortia Labs builds such assistants for customer service, sales and internal questions.

Request a first call AI consulting

AI chatbot
RAG
ChatGPT with your own data
AI knowledge base
Customer service chatbot

Your documentsManuals, contracts, FAQ
SearchFinds the right passages
Language modelWrites the answer
Answer with sourceTraceable for everyone
Your teamTakes over when needed

AI chatbot

What companies use a RAG chatbot for

Internal knowledge assistant

Staff ask about travel expense rules, product details or how a complaint is handled and get an answer from manuals, policies and the wiki, with a link to the original document.

Customer service on your website

The chatbot answers recurring questions from your product information and FAQs. When it finds no suitable source, it hands the conversation and its history over to your support team.

Answers with sources

Answers point to the documents they come from. Your team and your customers can check them, and the questions show you which documents are missing or out of date.

Permissions as in the original

We carry over the access rights of your sources: the chatbot only searches documents that the person asking may open in the original. HR documents therefore do not appear in answers for the sales team.

Vector search in your database

Documents are split into passages and stored as vectors, for example with pgvector in PostgreSQL or in Qdrant. New and changed files are picked up automatically, so answers match the current state.

Approach

From documents to a working chatbot

1
Review the sources
Which documents should be the basis, where are they, how current are they and who may read them? This inventory alone often reveals gaps that can be closed first.
2
Prepare and index
PDFs, Word files, wiki pages and tickets are extracted, split sensibly and made searchable. Tables and scans get their own treatment so that their content is kept.
3
Test with real questions
We collect questions your team and customers actually ask today and measure how often the chatbot finds the right source and answers correctly. Only then does it go live.
4
Maintain and improve
Unanswered questions and poor ratings go into a report. It shows which documents should be added and where search or instructions need sharpening.

From our own work

Software we build and run ourselves.

Screen Screen recorder for WindowsScreen, our screen recorder for Windows, has an optional AI chat that is switched off by default: it answers questions about the live transcript or about a screenshot. Omnia Document archive with AIIn Omnia, our document archive with AI, the AI suggests title, sender and date for new documents, and full-text search finds them again, also inside ZIP archives. A chatbot that has to find the right sources builds on exactly this kind of groundwork.

Questions about AI chatbot

What does RAG mean?

Retrieval-augmented generation: before answering, the system picks the relevant passages from your documents and passes them to the language model, which writes the answer based on them. Your documents are not trained into the model and can be changed or removed without retraining anything.

Can the chatbot give wrong answers?

Yes, language models can be wrong. RAG reduces that risk because the answer comes from your documents and names its source. We also set the chatbot up to say when it finds no matching passage instead of guessing, and we test with your real questions before launch.

Where are our documents and vectors stored?

Wherever your requirements point: in an EU data centre, on your own server or in your existing cloud. For highly confidential material the language model can run locally as well, so that neither documents nor questions go to an external provider.

Which file formats can the chatbot use?

Typically PDF, Word, Excel, PowerPoint, text and HTML, plus content from wikis, ticket systems or databases through their interfaces. Scanned documents are made readable with text recognition first. Which sources make sense is settled during the initial review.

Read on.

LLM integrationA language model is useful where your staff already work: in the ERP, the CRM, the ticket system or your own web application.AI document processingSupplier invoices, delivery notes, contracts and forms arrive as PDFs, scans or phone photos, and are often still typed in by hand.Local AINot every piece of information should go to a cloud provider: personnel files, engineering data, client records or health data are better kept in…PostgreSQLPostgreSQL is open, mature and can be extended for a wide range of tasks.AI agentsAn AI agent works through a task in several steps: it reads a supplier invoice, compares it with the purchase order and delivery note, checks the ERP…Speech to textMeetings, customer calls, dictation and voice messages hold information that rarely gets written down in full.

All about AI consulting All services and topics

Which questions should your chatbot answer?

Send us a few typical questions and tell us where the answers are today. The prototype shows how well a chatbot answers them from your documents.