Skip to content
AI for business

A custom AI chatbot and assistant (RAG)

We build a chatbot that answers from your documents and data via RAG, cites a source for every answer, and connects to your CRM, ERP, and helpdesk. For sensitive outputs, the final word stays with a person.

A custom AI chatbot and assistant over company documents (RAG)
18
years in business
4 h
response for retainer clients
NIS2
cybersecurity and backups
Brno
own infrastructure
  • Response within 4 hours
    SLA for retainer clients
  • NIS2 compliance
    Cybersecurity and backups
  • IT outsourcing
    Managed service and projects
  • Brno + 50 km radius
    On-site and remote
  • 18 years in the field
    Since 2008

Quick summary

A custom chatbot and assistant — an AI interface built over your documents and processes that answers with traceable source citations and can connect to your CRM and ERP. It is not a universal conversational tool, but a company specialist designed to pull data precisely from your environment and to leave the final word with a person for sensitive outputs.

  • Answers from your data with a link to the source — every piece of information is backed by a citation of a specific document, guideline, or ticket, so the output can be verified before it reaches a customer.
  • It understands your company, not general conversation — the model works with your terminology, products, and internal procedures, so it does not lose the context or improvise beyond its intended scope.
  • Connection to CRM, ERP, helpdesk, and the web — the assistant draws on current data in the systems you already use and can also write information back into them.
  • A human in the loop for sensitive answers — for drafts that require verification, AI prepares a proposal, a person confirms or adjusts it, and only then is it sent.

A chatbot that knows your company

A general chatbot answers based on public data and easily slips into hallucination as soon as it hits a specific company query. An assistant built on RAG, by contrast, looks for answers solely in your internal sources — in guidelines, technical manuals, contracts, the company wiki, or ticket history. For a query in Czech it selects the most relevant passage, assembles an answer from it, and attaches a link to the original document, so the operator or customer can see where the information comes from. This fundamentally reduces hallucinations and shortens the time that manual searching across several systems would otherwise take. The result is consistent communication that matches your internal standards, even in situations where the query concerns a narrowly specialised area.

Where a custom chatbot pays off

A practical deployment makes sense wherever queries on familiar topics recur and where the speed of the answer affects the customer experience or the workload of specialists.

  • A customer chatbot on a website or e-shop — it handles routine queries about availability, parameters, and order status and, if needed, passes the conversation to a live operator with a pre-filled summary.
  • An internal knowledge assistant for service, sales, and HR — a service technician asks about a repair procedure and gets an answer with a link to a specific manual; a salesperson finds the current terms of a contract, an HR officer the details of benefits.
  • Draft replies to e-mails and tickets — from the communication history and the knowledge base the assistant suggests a draft reply, which the employee checks, fine-tunes if needed, and sends.
  • Search across the document archive — contracts, technical drawings, or legal opinions become accessible by asking a natural question instead of trawling through the folder structure.
  • Help with onboarding and training — a new colleague asks about internal procedures and gets answers consistent with company guidelines, not a random interpretation from the colleague at the next desk.

How we deploy it

The approach is designed so that the first practical results are visible in a matter of weeks, not months.

  1. Analysis and selection of data sources and a suitable model — we map which systems and formats hold the relevant knowledge, and based on the nature of the queries we choose a model that makes sense in terms of speed, accuracy, and operating costs.
  2. Building RAG over your documents and setting up citations — we split the documents into semantic blocks, index them, and connect them to an answer-generation mechanism that always cites the source.
  3. Connection to systems — the web, CRM, ERP, helpdesk — and single sign-on — the assistant becomes part of the existing infrastructure, so it draws on current data and respects user permissions.
  4. A pilot on your data, evaluation, then operation and fine-tuning — on a limited group of users we verify the accuracy and benefit, adjust the configuration, and only then launch across the board with ongoing refinement.

Why ITHOPE

We are not resellers of generic AI packages — we build the solution on our own infrastructure and keep control over it from design to operation.

  • Our own GPU lab and hardware in Brno — computing power and development run on physical infrastructure that we manage ourselves, which for some plans means the data does not have to leave our or your controlled environment and you can more easily document an overview of its processing.
  • 18 years under one brand, IT and security — continuous practice in ICT management and cybersecurity means we do not deploy AI as an isolated experiment, but as part of a securely functioning whole.
  • One partner for IT, AI, and operation — from delivering server infrastructure through building the model to long-term support, you do not have to deal with several suppliers and their mutual shifting of responsibility.
  • We build and operate to an SLA — after deployment we ensure availability, monitoring, and incident response in a mode you set according to your needs.
  • A human in the loop, not a black box — we design the system so that decision-making authority stays on your side and you always know why the assistant answered the way it did.

Let’s arrange a consultation

Let us sit down over the specific documents and processes the chatbot could serve. We will go through which data it can draw on, where full automation pays off, and where manual control should be kept. We will fairly advise whether AI deployment in a company makes sense in your case, or whether it is wiser to start elsewhere.

If you are also interested in the economics, we have prepared an overview, How much AI costs for a company, to help you get your bearings in the topic before the meeting. Get in touch — we will arrange a date and prepare materials tailored to your situation.

FAQ

Frequently asked questions

How does a custom chatbot differ from an ordinary chatbot?
An ordinary chatbot answers from general knowledge and easily improvises on company queries. A custom chatbot draws via RAG solely from your documents, data, and processes and cites a source for every answer, so it responds in the context of your company and the output can be verified.
How does the chatbot prevent itself from making answers up?
It builds on RAG technology: the model first finds the relevant passages in your documents and only then assembles an answer, which it supplements with a link to the source. The risk of hallucinations drops significantly, and for unclear queries the model can be set to rather offer a handover to a person.
Can the chatbot be connected to our systems, such as CRM or ERP?
Yes. We connect the assistant to the systems you already use — the web, CRM, ERP, helpdesk, or intranet — via their interfaces. It draws on current data, respects user permissions, and, depending on the setup, also writes into the systems. We propose the scope and method of connection after analysing your environment, so as not to interfere with what already works.
Will company data stay secure?
We address data sensitivity from the start. The model can run locally on our hardware in Brno or directly in your environment, so the data does not have to leave for the public cloud. We set access rights by role, and for sensitive answers we leave the final approval with a person, not the model.
How long does it take to deploy a chatbot over our documents?
A first working pilot on your data is usually a matter of days to weeks, depending on the volume and quality of the materials. Only after verifying accuracy and benefit on a limited group of users do we launch the solution across the board, connect it to systems, and continue to fine-tune it. We will confirm timescales after reviewing your materials.
Can the chatbot answer well in Czech?
Yes, with the right choice of model and setup. Because it answers from your documents via RAG, it sticks to your terminology and the context of your field. We always measure the quality of answers first on real queries in a pilot, before deploying the solution into everyday operation across the team.
Will the chatbot replace our people?
No, we design it as support. It handles routine queries on its own, and for more complex ones it prepares a draft reply and passes it to a person for checking and sending. The aim is to save time on repetitive work, not to hand decision-making or responsibility to the model — those stay with your team.
Call Contact