unthinkable
Industries · You build a product

Your product was meant to give every learner the patience of a great teacher.

Students stall when help comes too late: a concept that won't click at midnight, an unanswered aid question, a missing document.

Build the product that is there at those moments, with teachers and staff making every decision that counts.

14.2%of first-year students did not return for springNational Student Clearinghouse, 2026
3 in 4calls to the federal FAFSA help line went unanswered, January to May 2024GAO, 2024
372students share each school counselor, against 250 recommendedASCA, 2024-25

The student's journey your product can change At every step of the student's journey.

What changes for each role (text version)

Applying

  • What if applicants knew within a day what was missing from their file? Today: Common App processed 10.8 million applications in 2025-26, 7.06 for every applicant. With an agent: Each file is checked on arrival, and applicants are told exactly what is missing and when it arrives. Rule, not model: What each program requires is a checklist in code, not the model's opinion. (Common App End-of-Season Report, 2026)

The decision

  • What if you heard your decision the day it was released, and got answers before the deposit deadline? Today: Applicants check portals for weeks, then wait days for answers about scholarships, deposits and housing. With an agent: Applicants hear the moment their decision is released and get answers from their award letter and approved content. Rule, not model: Release dates and award terms come from the admissions system and the award letter, never from the model.

Financial aid

  • What if a financial aid question got an answer at 11 pm, not a busy signal? Today: Nearly three-quarters of calls to the federal FAFSA help line went unanswered from January to May 2024. With an agent: Students get answers from aid policy and their own record at any hour, and aid staff handle what needs judgement. Rule, not model: Aid amounts, missing items and holds are read from the systems of record, never worked out by the model. (US Government Accountability Office, 2024)

Paying

  • What if parents could pay and plan without calling the bursar? Today: Parents call the bursar during office hours to find out what they owe and how to split it. With an agent: Balances and payment plans are explained and set up at any hour, within what the student has shared. Rule, not model: Plan terms and fees come from the bursar's published rules.

Registering

  • What if a full class didn't mean a week of emails to get in? Today: When a required class is full, students email departments and advisors and wait to hear back. With an agent: Students see open sections, waitlists and prerequisites at once, and override requests reach the right person as a complete case. Rule, not model: Seats, prerequisites and degree requirements are checked by the SIS and degree audit, not by the model.

Help during term

  • What if a struggling student was pointed to help the moment they asked? Today: A student who reaches out late at night may not speak to a person until the next day. With an agent: Crisis contacts appear straight away, and a counselor is alerted to reach out the same day. Rule, not model: The wording and contacts in a crisis come from counseling's approved script.

Staying on track

  • What if a student who was slipping heard from someone in week 4, not after midterms? Today: 14.2% of students who started college in fall 2024, about 595,000 people, did not return for spring. With an agent: Students who match the early alert rules get a check-in from their instructor and an advisor meeting within days. Rule, not model: Who gets a check-in is decided by the university's early alert rules, not by a model's prediction. (National Student Clearinghouse Research Center, 2026)
What your users get

What learners, teachers and hiring teams get.

AI tutor

Guides students through problems step by step using your curriculum, without giving away answers.

Feedback and grading assistant

Drafts rubric-based feedback on written work for the teacher to review and release.

Course and content builder

Turns source material into lessons, quizzes and practice sets aligned to learning objectives.

Recruiting copilot

Screens applications against stated criteria, summarises evidence and drafts outreach for a recruiter to approve.

Interview scheduling and candidate agent

Answers candidate questions and books interviews across calendars.

Employee learning coach

Recommends training based on role and goals and answers policy questions.

How it becomes real

Your product helps every learner the moment they're stuck, and teachers and staff make every grade, admission and hiring decision.

Added to the product you have

Into the product your users already know, released in steps, without breaking what they rely on.

Or a new product, built with you

From an idea to a first release, with a complete team from the first week. You own the code and the IP.

The first stepFix one moment in your users' journey, measured.

We pick the moment where your users wait longest, fix it first, and measure it before and after: time to an answer, how many drop off, how many come back.

Why you can count on us

Already senior. Already on the job.

What we have built, run and secured. Each line is something we can show you in the first call.

  • Campus systemsStudent information systems, admissions CRMs (Slate, Salesforce) and LMS, with FERPA and GDPR handling.
  • Regulated dataHIPAA with a BAA chain down to the model provider, GDPR with EU hosting, no training on your data. See it →
  • Tested before customers see itEvery AI step ships with an evaluation set and pass thresholds agreed with you; a release below threshold does not ship. See a sample report. See it →
  • Every action on the recordEach agent action is logged with what it read, the rule that applied and who approved, so it can be replayed for an auditor. See it →
  • A senior team from week oneA senior lead, designer, engineers, QA and DevOps. No months of hiring first.
  • Products we run for yearsThe team that builds a product keeps it running: releases, monitoring and on-call after launch.
  • Yours at handoverCode, prompts, evaluation sets, runbooks and dashboards in your accounts, so your team can run it without us. See it →
  • Client referencesAbout 20 client testimonials on our current site, 12 from US clients and 2 from the UK. Source →
Your product, your IP.

Code, models, prompts and data are yours. No platform lock-in and no licence to rent.

Ship before you hire.

Design, AI, architecture, testing and DevOps from day one. Hire your own team when it makes sense, with a full handover.

For the technical reader

Where products in this space get stuck, and how we handle it.

Student and candidate privacy

We design for FERPA, COPPA where minors are involved, and GDPR, keep data in the region you need, and never train models on learner or candidate data.

Pedagogy and accuracy

Tutors draw on your approved content, are tested on real student questions with your educators, and are tuned to guide rather than give answers.

Fair hiring

Screening uses job-related criteria set by people, is tested for adverse impact, and supports bias audits such as NYC Local Law 144 and EU AI Act high-risk duties.

Scale and cost

Usage spikes around terms and hiring seasons. We control cost per learner or per applicant with caching, smaller models and usage limits.

Who we build with

You know the problem. We bring the complete team.

  • Teachers, trainers and learning designers turning a method into a product
  • CTOs of ed-tech, learning and HR-tech platforms adding agents
  • Recruiting and talent leaders building their own hiring tools
You inherit

Building blocks already built.

  • Curriculum-grounded retrieval
  • Rubric-based feedback engine
  • Age and role-based access controls
  • Candidate evaluation with evidence and audit log
  • LMS and ATS connectors (LTI, Canvas, Greenhouse, Workday)
  • Evaluation suite for accuracy and fairness
Questions buyers ask

Straight answers.

How do you build an AI tutor that does not just give students the answers?

We ground the tutor in your curriculum and set its behaviour to guide: ask a question back, give a hint, check understanding, then move on. We test it with educators on real student questions, including attempts to get the answer directly, and score each response. Teachers can see conversations and adjust the approach. Accuracy and teaching quality are both measured before launch.

Is using AI with student data FERPA compliant?

It can be. We use model providers that do not retain or train on your data, minimise what personal data reaches the model, and keep records in the region and account you choose. Access is role-based and logged. For under-13 users we design for COPPA. We document the data flow so schools and districts can complete their own privacy review.

Is it legal to use AI to screen job candidates?

Yes, with care. In New York City, automated hiring tools need a bias audit and candidate notice, and the EU AI Act treats recruitment AI as high-risk, with duties for documentation, oversight and accuracy. We build screening that uses job-related criteria set by people, shows the evidence for each recommendation, keeps a recruiter in the decision and supports independent bias audits.

How much does it cost to build an AI tutoring or learning app?

It depends on how much content the app covers, which platforms it runs on and its integrations, such as LTI or Canvas. A first release is usually one fixed outcome, and you get a written estimate after a short discovery. Running costs depend on how much students use it, so Unthinkable measures cost per learner during that first release and sets limits per plan, keeping costs predictable through exam season.

I am a teacher with an idea for an education product. Can you build it?

Yes. You bring the teaching method and the problem; we bring the complete product team, from product designer to AI engineer, testers and a senior lead. We start with a short discovery that turns your idea into a scoped first release with a fixed price. You own the code, IP and data, and we stay until it works in real classrooms.

Can you add AI features to our existing LMS or ATS product?

Yes. We work in your codebase and add agents behind your existing permissions, such as feedback drafting, content generation, candidate summaries or scheduling. We connect to Canvas, Moodle, Greenhouse or Workday where your product relies on them. Each feature ships with tests and cost tracking, and your team owns and can extend it after handover.

Ask Unthinkable

Which moment would you fix first?

Tell us where your users wait today. Our AI assistant, Ask Unthinkable, helps you define your requirements, then connects you with the right team.

Send us a note instead

Required

Thanks. A senior AI engineer will reply to you.
u
Ask UnthinkableClarity first, then the right team
Your conversation

An AI agent. This prototype uses prepared answers and sends nothing.