Skip to content
Prospicience Consultancy
E14In production

An admissions assistant that answers in Gujarati, Hindi or English, by text or voice

Pre-school group

For a pre-school group, Prospicience built an admissions assistant that sits on every page of the school site and answers parents in Gujarati, Hindi, English and other Indian languages, typed or by voice note. Questions it cannot answer from the school's confirmed facts become leads that are saved, emailed to admissions and confirmed back to the parent.

SpeedBuilt and live in five days

languages answered
Gujarati, Hindi, English and more
ways to ask
typed or voice note
voice samples on this page
5
lead delivery
saved, emailed and confirmed
Five voices from the assistant, in three languages
  • English, voice one

  • English, voice two

  • English, voice three

  • Gujarati

  • Hindi

The challenge

Admissions enquiries arrived by phone and message at all hours, in three languages. The team answered the same questions about fees, timings and transport again and again, and genuine callback requests got lost in between.

Why it mattered

Every lost callback is a family that may enrol somewhere else, and admissions season is exactly when the team has the least time to chase them. Answering routine questions by hand also kept staff away from the conversations that turn a visit into an enrolment.

What we built

  • Put an assistant on every page of the school site that answers in English, Gujarati, Hindi and other Indian languages.
  • Lets parents type or send a voice note, and hear the reply spoken back in the same language.
  • Turns questions outside the answer bank into leads: the assistant collects the details, saves them, emails the admissions team and confirms back to the parent.
  • Gives the team an admin view to work through every enquiry, read the full conversation behind it, and switch the AI model for every visitor without a redeploy.

The result

  • Parents get an answer at any hour, in the language they speak, on any page of the site.
  • Every enquiry the assistant cannot close lands in three places: the database, the admissions inbox and the admin view.
  • Spoken replies are generated in about a third of a second in English and Hindi.
  • The team can move to a better or cheaper AI model from the admin view, live, with no developer involved.

Built for trust

The assistant quotes only figures the school has confirmed, and anything outside its answer bank becomes a hand off to a person rather than a guess. Each service degrades gracefully: if voice is unavailable, the microphone button simply disappears and text chat carries on, so a parent never meets an error.

AI at work

AI answers parents on every page of the school site, day and night, in Gujarati, Hindi, English and other Indian languages, typed or by voice note. Every request it cannot settle becomes a lead the admissions team picks up.

What it produces

  • Answers in the parent's own language
  • Spoken replies to voice notes
  • Captured leads with the full conversation
  • Emails to admissions and confirmations to parents

How it works, day to day

A parent types a question or sends a voice note. AI answers from the school's confirmed facts in the same language, with a spoken reply a tap away. When a question falls outside those facts, it takes the parent's details, and the admissions team follows up with the whole conversation in front of them.

Where it saves time, and gets better with use

  • Routine questions on fees, timings and transport are answered at any hour, so the team's time goes to families ready to visit.
  • Spoken replies are ready in about a third of a second in English and Hindi.
  • A fact the school confirms is added to the answer bank once and used in every conversation after, and anything missing becomes a callback rather than a guess.
  • The team can switch to a better or cheaper AI model from the admin view as models improve, live, with no developer involved.

By the numbers

Gujarati, Hindi, English and more
languages answered
typed or voice note
ways to ask
5
voice samples on this page
saved, emailed and confirmed
lead delivery
4, chosen per key
AI providers behind it
0.30s English, 0.32s Hindi, 0.75s Gujarati
voice reply generation, measured
about 400
AI models selectable from admin

The technology

  • Answers on every page of the school site
  • Type a question or send a voice note
  • Replies in the same language the parent used
  • Turns a real enquiry into a lead the team sees
  • Keeps working when a service is unavailable
  • Claude (Anthropic)
  • GPT (OpenAI)
  • Gemini (Google)
  • ElevenLabs
  • Sarvam
Technical notes
  • Multiple model providers behind one interface, with the admin listing every model the configured keys can reach, read live from each provider catalogue
  • Speech recognition and speech synthesis through an Indic speech provider alongside a general voice engine
  • Per key graceful degradation: absent capability is hidden in the interface rather than surfaced as an error
  • Leads persisted to Postgres, with an in memory fallback so the assistant still answers if the database is unreachable
  • Transactional email to the admissions team on every captured lead
  • A published system prompt defining refusal, escalation and the hand off contact

Questions we get asked

Can an AI assistant answer parents in Gujarati or Hindi?
Yes, by text or by voice note, and the reply comes back in the language the parent used. Voice samples in English, Gujarati and Hindi are on this page so you can hear the quality for yourself.
What happens when a parent asks something the assistant does not know?
It collects their details, saves the lead, emails the admissions team and tells the parent someone will get back to them. It never invents a fee or a timing, so the school never has to walk back a promise.
Where do the enquiries end up?
In a database, in an email to the team, and in an admin view they can work through, each with the full conversation attached. Different people check different places, so every lead reaches all three.

Related systems

Have a process that should run itself? Let us build it.

Tell us the problem. We will show you what we would build, how fast we can ship it, and what it will deliver, from a first pilot to a full platform.