An AI chatbot for higher education that answers only from your own sources
HiWi+ Chat answers questions from prospective and current students around the clock: on the study programme, application, admission, enrolment, re-registration and study organisation. Exclusively from the pages and documents of your institution, with a source reference for every answer.
The questions that come back every semester
By far the largest share of enquiries in offices and student advisory services are always the same ones. Those are exactly what the chat takes over, without anyone retyping a line of the answer.
Programmes and course offering
Content, prerequisites, standard period of study, specialisations and differences between related programmes, answered from your own pages and leaflets.
Application and admission
Deadlines, documents, admission restrictions, procedures for international applications. The most frequent question before the semester and the one that costs the most time.
Enrolment and re-registration
Which documents are still missing, by when re-registration is due, what happens if a deadline is missed. Answers around the clock, including the weekend before the cut-off.
Start of studies and organisation
Semester schedule, registering for courses and examinations, leave of absence, changing subject, recognition of credits from earlier studies.
Study and examination regulations
The chat reads the version of your regulations and module handbooks currently in force and names the passage each answer comes from.
Certificates and formalities
Semester fee, semester ticket, enrolment certificate, leave of absence, de-registration. Small questions in large numbers that run through the office all day.
Questions in English and other languages
International applicants ask in their own language and receive the answer in that language, based on the same German sources. Without your pages having to be translated for it.
Referral to the right place
Anything that is not documented or needs a personal decision leads to a checked referral to the responsible office or student advisory service, not to a guessed answer.
How an answer comes about
The difference from a general language model is not the model but what it is allowed to see. Four steps, each of them traceable.
Classify the question
The chat first checks whether the question falls within its remit. Anything outside is declined before a model even answers.
Find the matching passages
From your knowledge base the passages that match in substance are retrieved, not whole documents. What counts is the meaning of the question, not the keyword searched for.
Phrase the answer only from those
The language model may use only the passages found. World knowledge is switched off. If the sources do not support an answer, the chat says so and refers on.
Supply the source
Every answer names the sources used. Your editorial team additionally sees which passages in particular were drawn on.
Your content, kept current
What is connected are your institution's web pages as well as PDF, Word and Excel files. Study and examination regulations, module handbooks, leaflets, deadline overviews and the question collections that have grown in the offices.
On your website, in half an hour
Widget
Two lines in the page template are enough. The widget is encapsulated, adopts your colours and your greeting text and loads nothing from third-party servers.
TYPO3 and other systems
In TYPO3 via Fluid, TypoScript or a content element of your own, with a data protection gate to your specification. Equally in WordPress, Drupal or a custom-built site.
A chat page of its own
Its own address under your domain, linkable from a newsletter, a notice or a QR code. Useful for prospective students who are not yet on your website.
API
For your own interfaces the programming interface is available, with the same session and abuse logic as the widget. That lets you build the service into a student portal as well.
Four routes, depending on how strict your requirements are
By default neither a sign-in nor personal data is needed to use it. Optionally a sign-in via your single sign-on can be placed in front, without a second data store arising in the platform.
Operated in Germany
We operate the platform in a German data centre. No effort for your IT, ready to go immediately. The usual route for getting started and for a pilot.
German operation, European models
The same operation, but you determine which model providers are permitted at all. Anyone who does not want requests to US providers restricts the choice to European models.
Your own data centre
The same platform as a container in your own environment, operated by your IT. Billed as an annual licence rather than monthly operation.
Fully local AI server
With HiWi+ Local AI the language model also runs on your own hardware. Air-gap capable, without any request leaving the network.
A data processing agreement, a description of the technical and organisational measures and a text block for your record of processing activities are part of the setup and not an extra item.
You keep control over every answer
Test catalogue before every release
A catalogue of real questions runs automatically against the current state before it goes live. The result is a log you can present.
Question list from live operation
The editorial team sees what is actually being asked and where answers are missing. That produces a task list for your website, not only for the chat.
Hard boundaries instead of hope
Limits of remit, the offices that may be named and the handling of cited paragraphs are stored as rules and not left to the model.
Knowledge spaces per area
Every faculty maintains its own sources and its own editorial team, on a shared platform with jointly maintained general content.
What the chat deliberately does not do
A tool that promises everything is measured by its first wrong answer. That is why we say in advance where the boundary lies.
Karlsruhe University of Applied Sciences
HiWi+ Chat is in use at the Faculty of Business Administration of Karlsruhe University of Applied Sciences and was developed there together with the departments. Seven programmes, their regulations and module handbooks as the knowledge base, tested against a catalogue of real questions from student advisory services.
“Our main goal was to reduce administrative processes and workload within the faculty while at the same time offering students a better service, for example with the faculty chatbot they can ask questions about their studies around the clock. The HiWi+ modules were developed for our specific needs at the faculty, but they are certainly just as usable for many other educational institutions.”
Frequently asked questions
Because it may answer only from your approved sources and world knowledge is switched off. Every answer names its source, and your editorial team sees the passages used. If the sources do not support an answer, the chat refers to the responsible office instead of guessing.
Your institution's web pages, PDF documents such as study and examination regulations or leaflets, as well as Word and Excel files. Content is separated thematically into knowledge spaces, so that one faculty does not answer from another's documents.
As a widget via two lines in the page template, encapsulated and without loading anything from third-party servers. In TYPO3 it can be integrated via Fluid, TypoScript or a content element of your own, taking around half an hour. Alternatively as a chat page of its own or, via the programming interface, inside a student portal.
Yes. By default neither a sign-in nor personal data is needed to use it. Operation runs either in a German data centre, in your own data centre or entirely locally on your own hardware. No training takes place on your data. A data processing agreement, the technical and organisational measures and a text block for your record of processing activities are part of the setup.
Your editorial team, through an administration interface that needs no technical background. Sources can be replaced, added to and given expiry dates. For the first semester we reckon with around two hours per week, considerably less after that.
A pilot at faculty level is usually productive within six to ten weeks, including capturing the sources, the test catalogue, data protection documentation and training the editorial team. A free demonstration phase with your own documents comes before that.
Billing consists of a one-off setup and monthly operation per stage of expansion. There are no per-user licences, usage is unlimited. A pilot at faculty level starts in the mid four-figure range and is credited in full against a later expansion. For operation in your own data centre there is an annual licence.
See the chat working on your own documents
Before any engagement there is a free demonstration phase. You provide your own regulations and leaflets and check on real questions how robust the answers are. Without commitment and at no cost.