Multilingual AI assistant for international students
A 24/7 assistant for international students and staff: admissions, housing, visas, campus life. Trained on your own documents, answering in their language.
A multilingual AI assistant for universities is a chat assistant trained on an institution's own admissions, housing, visa and campus documentation, which answers prospective and enrolled international students in their own language at any hour. HumAIn Connection builds these in more than nine languages with human quality assurance on the highest-volume ones.
The problem with the international student inbox
International admissions runs on repetitive questions asked in many languages at hours when nobody is at a desk. What documents does my application need. Is my prior degree recognised. When does housing open. Do I need a residence permit and how long does it take. What happens if my visa is late.
Each individual question is cheap to answer and the volume is what breaks the team. The questions arrive in the applicant's language, the office answers in English or the national language, and the gap in between is where applicants quietly give up. Applicants who go silent are usually recorded as not interested, when the honest description is that they could not get an answer.
Generic chatbots make this worse rather than better, because a chatbot that does not know your specific deadlines confidently invents them. In admissions, a hallucinated deadline is not a support failure. It is a student who missed an intake.
What we deliver
The assistant is grounded in your documentation, not in a general model's memory of how universities usually work.
- Trained on your own documents
- Your admissions pages, housing rules, visa guidance, programme handbooks and FAQs. When the answer is not in your material, the assistant says so and routes to a human instead of guessing.
- Web widget and WhatsApp
- The site widget covers prospective students researching you. WhatsApp covers the ones already in transit, which is where the urgent visa and housing questions actually arrive.
- Answers in 9+ languages, around the clock
- Students ask in their own language and get an answer in it, at the hour they are awake rather than the hour your office is open.
- Human QA on your top student languages
- Professional translators review the assistant's real answers in your highest-volume languages and correct the source material where the output is weak, so quality improves against real traffic rather than against a test set.
- Escalation to a person
- Anything sensitive, ambiguous or outside the documentation is handed to your staff with the conversation attached, so the student does not repeat themselves.
How a project runs
Same four steps, and the review step is continuous here rather than one-off, because an assistant keeps answering after launch.
Listen
We work out which questions actually dominate your inbox, which languages your applicant pool really uses, and where the authoritative version of each answer lives. That last one is usually the hardest question to answer internally.
Build
We ingest your documentation, build the assistant, and connect the web widget and WhatsApp. Deliberate refusal behaviour is configured here: what it must never attempt to answer.
Review
Before launch, professional translators test the assistant in your priority languages against real historical questions. After launch, they keep reviewing real conversations.
Care
Deadlines change every cycle. We keep the source material current, and the assistant changes with it rather than confidently repeating last year's dates.
What the human review step actually involves
For an assistant, review is not a one-time sign-off on a deliverable. It is an ongoing audit of what the system is telling real students.
- Real conversations are sampled and read by a native speaker of the language the student wrote in.
- Answers are checked for factual correctness against your source documents, not just for fluency. A confident, well-written wrong deadline is the failure mode that matters.
- Refusal behaviour is tested: the assistant must decline to answer immigration questions it cannot ground in your material, rather than improvising.
- Tone is checked per language, because directness that reads as helpful in Dutch can read as abrupt in other languages.
- Gaps found in review are fixed in the source documentation, which improves the assistant and your website at the same time.
What it will not do
It does not give immigration advice. It points at your institution's official guidance and at the relevant authority, and escalates anything beyond that to a human. Automated immigration guidance is a liability that no university wants and no vendor should be selling.
It does not replace your admissions team. It absorbs the repetitive volume so that the team spends its hours on the cases that genuinely need a person, which is the only part of the job that was ever a good use of their time.
Timeline
| Scoping and documentation audit | 1 to 2 weeks |
|---|---|
| Build and ingestion | 2 to 3 weeks |
| Pre-launch human QA in priority languages | 1 week |
| Pilot running against real traffic | Typically one intake cycle |
What drives the cost
We do not publish a rate card yet, because a number without a scope behind it is noise. What we can tell you is exactly what moves it, so you can size the work before you talk to us.
- Number of channels: web widget alone, or web plus WhatsApp.
- How many languages get human QA rather than AI-only coverage.
- The state of your source documentation. Scattered, contradictory or out-of-date material is the single largest cost driver, and the audit usually pays for itself.
- Integrations with an existing CRM, student information system or ticketing tool.
- Ongoing maintenance scope, since deadlines and rules change every cycle.
Every engagement starts with a scoped pilot, priced and agreed in writing before any work begins. No open-ended retainers, and no per-word invoicing that nobody can forecast.
Questions we get asked
How do you stop it inventing deadlines?
The assistant answers from your documents rather than from the model's general knowledge, and it is configured to say that it does not know and route to a human when the answer is not in your material. We test that refusal behaviour explicitly before launch, because it matters more than the answers it does give.
Which languages get human review?
The ones your applicants actually use in volume, which we identify during scoping rather than assuming. Other languages are still answered, but they are covered by the AI pipeline without a continuous human audit, and we are explicit with you about which is which.
Where does the student data go?
This is a question your data protection officer should be asking, and we answer it in writing with the specific subprocessors involved before any pilot starts. We will not make a blanket claim on a web page about a setup that has to be configured per institution.
Can it hand over to our staff?
Yes, and it should. Anything sensitive, ambiguous or outside the documentation is escalated to your team with the full conversation attached so the student does not have to start again.
What happens when our deadlines change?
The source material is updated and the assistant follows. This is the part most institutions underestimate: an assistant is not a project that finishes, it is a system that has to stay current, which is why maintenance is part of the engagement rather than an afterthought.
Related services
Start with one pilot, and judge the review.
Tell us what you are trying to do and we will tell you honestly whether we are the right people for it.
Prefer email? hugo.megardon@megaaisolutions.com