UFITProduct Hub

Integrated overview · for boards, investors and policy makers

Two axes of a child’s learning. One principle.

Science and writing are where a primary-school child’s thinking becomes visible. UFIT applies one measurement principle to both — Aprendiz Inteligente for how a child reasons through science, VeriWright for how a child builds a sentence. Neither scores only the finished product. Both read the process that produced it, and turn it into something a teacher can act on this week — and, where it matters, early enough to change the outcome.

🇧🇷 CELEPAR — approved public-tender supplier (Paraná State ICT company) 400,000 users in Paraná 🇧🇷 Paraná & Federal Gov. partnership 🇵🇾 Paraguay 300+ classrooms (China & S. Korea) KCA President’s Award 2025 · Digital Innovation

The framework

One principle, two axes, one profile of the child.

A school already knows what a child produced. What it rarely knows is how — where the thinking went fast, where it stalled, where it went back and repaired itself. Both products capture that, on their own axis, and write into the same profile.

Axis A · Science

How a child reasons through a problem

Aprendiz Inteligente · in service in Brazilian public schools

  • LearnAn AI learning board builds the lesson from approved textbooks only — concept, glossary, comparison tables, checks for understanding.
  • DoExperiment and investigation: a hypothesis written before the result is known, a variable the student manipulates, a conclusion the student must argue. Plus AI-rendered visuals and manipulable simulations for what a classroom cannot show.
  • AssessItems in the state assessment format, each tagged with its curriculum skill and its Bloom level — so a wrong answer names the idea to reteach, not just a lost point.
Axis B · Writing

How a child builds a sentence

VeriWright · controlled pilots, primary and secondary

  • CaptureA digital pen on ordinary paper records timing, pauses, pressure and pen movement. No tablet, no special surface, nothing about the lesson changes.
  • ReplayThe session is reconstructed — not the finished text, but the order and rhythm in which it appeared.
  • ReadThe trace is mapped onto the cognitive process behind it: planning, translating, monitoring, revising — and what each cost the writer.

One individual learning profile

Per child, over time, and read against that child’s own history rather than against a class average. Two subjects, one picture of how this learner works.

01

Action a teacher can take this week

Not a score to file. A named strength to build on, a named gap to close, and the specific next activity for each — for every child in the class, from one report.

02

An earlier intervention window

Process signals diverge before the product does. A child compensating hard to produce acceptable work is invisible to outcome-only marking — and visible here.

03

Inclusion without separate tracks

One common activity for the whole class, adapted by need rather than by diagnosis label. No child is sent to a parallel worksheet to be included.

Research foundation & collaboration

Built on established science, not on assertion.

Both axes rest on published research lines with named institutions behind them — the interactive-simulation pedagogy of a Nobel laureate’s project at the University of Colorado Boulder, and the German digital-pen research line at KIT, Fraunhofer IIS and DFKI.

A modern university research institute building with a glass and concrete facade.
The research baseUniversity institutes, not a vendor white paper. Each claim below traces to a named group with published work behind it.
A bright, well-kept classroom with rows of wooden desks and a green board, lit by daylight.
The settingOrdinary public-school classrooms — paper, pens and the timetable a school already has. Nothing about the lesson changes for the data to exist.
Close-up of a sensor pen tip resting on a circuit board on an electronics workbench.
The instrumentA pen with inertial sensors on its own embedded hardware — the line KIT, Fraunhofer IIS and DFKI publish on, and the signal VeriWright reads.

Images are rendered illustrations of the settings described — not photographs of the institutions named below, and not of any real classroom or student.

CU Boulder

Interaction as the learning mechanism

PhET Interactive Simulations · University of Colorado Boulder

Founded in 2002 by Carl Wieman, recipient of the Nobel Prize in Physics in 2001. HTML5 simulations across physics, chemistry, biology, earth science and mathematics — the premise being that a student learns a system by changing it and watching it respond, not by reading about it.

In AprendizThe pedagogical foundation of our simulation library — manipulable activities tied to the Brazilian curriculum.

phet.colorado.edu ↗

KIT

Writing captured on ordinary paper

Karlsruhe Institute of Technology · ITIV — project KIHT

Kaligo-based Intelligent Handwriting Teacher, a Franco-German joint project building an intelligent handwriting-learning device around a sensor pen. Inertial sensors reconstruct the writing trajectory with no tablet and no special surface; ITIV’s part is running those AI algorithms on the pen’s embedded hardware.

In VeriWrightPaper-first capture. The classroom does not have to change for the data to exist.

itiv.kit.edu ↗

Fraunhofer

The sensor signal, benchmarked

Fraunhofer Institute for Integrated Circuits (IIS) · OnHW dataset

Online Handwriting Recognition from Sensor-Enhanced Pens — raw time-series from a multi-sensor digital pen released as a public benchmark: triaxial acceleration, gyroscope, magnetometer and force at 100 Hz. The OnHW-chars set alone holds 31,275 letters from 119 writers.

In VeriWrightThe signal model our features are computed from, and a public benchmark to test against rather than a private one.

iis.fraunhofer.de ↗

DFKI

Cognitive state from pen signals

German Research Center for Artificial Intelligence · Interactive Machine Learning Lab

Two lines meet here: whether low-level pen signals can predict task difficulty and performance in primary-school children, and a style-versus-content neural architecture that confirms, passively and continuously, that the person writing is the person enrolled.

In VeriWrightThe cognitive reading behind the teacher report — and an integrity signal about authorship, not about detecting an AI.

iml.dfki.de ↗

From a pen stroke to an intervention

The chain a policy maker should be able to follow end to end. Nothing else is recorded, and nothing is inferred that does not start here.

1 · Raw streams
  • Triaxial acceleration
  • Gyroscope · rotation
  • Magnetometer · orientation
  • Force · tip pressure
  • Timing · ~100 Hz
2 · Derived parameters 150+

Cognitive-process parameters computed from those streams — pause length and placement, in-air time, burst structure, velocity variability, pressure stability, retrace and revision events.

3 · Machine learning

Models trained on those parameters estimate the writing process stage and its cost, and compare a session with the same child’s earlier sessions rather than with a class average.

4 · Detection & early intervention
  • Process profile per child
  • Class-level report for the teacher
  • Early flag → professional assessment
  • Structured plan for home

The 150+ figure is UFIT’s own feature count for the VeriWright pipeline. The institutions named above are cited as the published research basis of the approach; they are not an endorsement of UFIT, and a shared benchmark or sensor model does not transfer any published accuracy figure to our products.

Axis A · Science · in service

Running today in Brazilian public schools.

Science is the axis already in service — years 5, 7 and 8 — and it is also the proof that the engine is not subject-specific. The same six-tab lesson structure, the same generation-and-review pipeline and the same accessibility contract have already been carried into other subjects.

in service in development roadmap

One lesson engine six lesson tabs · generation + review accessibility contract · curriculum lock built once Science years 5 · 7 · 8 · in service Arts in development Geography / History in development Further subjects roadmap Adding a subject adds approved sources and activity types — it does not rebuild the platform.
Why the second subject is cheap. The board is a structure, not a syllabus. Science paid for the engine; every subject after it inherits the same six tabs, the same generation-and-review pipeline and the same accessibility contract.
Subject and year coverage. “In development” means active build work — prototypes exist on live open data, but the subject is not yet in service.
Subject5th year7th year8th yearBasis
Science ● in service ● in service ● in service Approved textbooks, curriculum-locked (RAG)
Arts ◐ in development Public-domain masterworks, gigapixel study
Geography / History ◐ in development National satellite & census open data, heritage field trips
Further subjects ○ roadmap — same engine, new content Curriculum sources per subject

How the AI board and the quiz items are actually made

The engineering behind “AI-generated”. Nothing is written from the open internet, and nothing reaches a student without a human signature.

01 · SOURCES Approved material state textbooks · BNCC curriculum 02 · INDEX Curriculum lock retrieval (RAG) with page-level sourcing 03 · GENERATE Board + items 6 lesson tabs · Bloom + distractors 04 · AUTO-CHECK Machine gates curriculum match · answer key · a11y 05 · HUMAN GATE Sign-off teacher + subject specialist, before use In class · on paper (OMR) or on screen results by curriculum skill and Bloom level feed the next lesson and the next quiz a failed check returns to 03 — it never skips 05
Two gates, in this order. Machines check what machines can check — that every item is on-curriculum, that the key is right, that the accessibility contract holds. What remains is judgement, and that is a named teacher and a subject specialist. The same pipeline produces the arts and geography boards; only the sources change.

Why expansion is cheap

The board is a structure, not a subject: objective, concept, experiment, investigation, exploration, glossary. Adding a subject means adding approved sources and activity types — not rebuilding the platform.

Simulations, by type not by title

61 science activities were classified into 5 reusable activity types; the three built so far already cover 46 of them. Types transfer across subjects and years.

Every child, one activity

From general classes to students with intellectual disabilities, sensory disabilities and neurodivergence — the same lesson, adapted by need. Support is switched on by need, not by a diagnosis label, and every AI-generated adaptation is reviewed by a teacher and a subject specialist before a student sees it.

One activity same science, same answer key Standard view Easy-read · same concepts Braille & tactile graphics Sign language (LIBRAS) first Switch, keyboard or pointer Same objective same class, same lesson
Adaptation, not segregation. Five ways into one activity — chosen by need, never by a diagnosis label — all landing on the same objective. The alternative that schools usually reach for, a parallel worksheet for “those” children, is the thing this design exists to prevent.
Axis B · Writing · primary and secondary

Composition assessed as a process, not only as a product.

Automated essay scoring reads what survived onto the page. It cannot see the twenty minutes that produced it. VeriWright assesses the writing process — and that difference is what makes early identification possible.

writing short pause long pause · searching pen in the air revision

Child A fluent long bursts · pauses fall at clause boundaries · no revision needed Child B under strain revision short fragments · long searching pauses inside phrases · pen hovering · one repair 0 min 14 min Schematic illustration — not a real student record.
Both children handed in an acceptable paragraph. On the finished page they are indistinguishable — and an outcome-only mark will treat them as equivalent. The difference is entirely in how the text was produced: Child B is compensating, paying far more to reach the same result. That is the child who is missed for years, and the reason the process is worth measuring.

Product-only assessment

What a finished text can tell you

  • Spelling, grammar, structure, length
  • Whether the task was completed
  • Nothing about effort — a child who struggled for forty minutes and a child who finished in five look identical
  • Nothing about a child compensating — producing acceptable work at an unsustainable cost
  • Nothing until the output is already failing — which is usually years late

Process assessment

What the production of the text adds

  • Where planning happened — before writing, or not at all
  • Where language came easily and where it was searched for
  • Whether the child rereads and repairs their own text — the mark of a self-monitoring writer
  • What the work cost, measured against the same child’s earlier sessions
  • Divergence visible while the product still looks acceptable

One class, one report, three conversations

The teacher receives the whole class in a single view, sorted into three bands — each with a different action attached. The point of the report is not the ranking; it is that every band gets something.

Process diverges early flag → assessment Typical range strengths named, gaps supported Strong writers extended, not parked Schematic — band widths illustrate the report structure, not measured prevalence.

Process diverges

Flagged early, referred properly

Where the process looks unlike the child’s peers and unlike their own baseline. The flag opens a professional pathway — it never closes one.

  • Early screening flag with the evidence attached
  • Referral to a qualified professional or specialist service
  • A structured intervention plan shared with the family and the child
  • Re-measurement against the same baseline to show whether it worked
Delivered as a referral pack: evidence, plain-language summary for parents, and the plan.

Typical range

Strengths named, one gap closed

The majority, and usually the group that receives the least specific feedback. Process data gives each of them something concrete.

  • One strength to keep, stated plainly
  • One gap to close, with the activity that closes it
  • Progress measured against their own earlier sessions
Delivered as a one-line action per child, teacher-editable.

Strong writers

Extended, not parked

The most commonly wasted group. Finishing early is treated as a reason for enrichment that builds on the specific strength the process shows.

  • The strength is named — planner, fluent drafter, disciplined reviser
  • An extension plan that stretches that strength
  • Peer roles where the strength is useful to others
Delivered as a per-child enrichment plan inside the class report.
A flag is a request for assessment, never a diagnosis. VeriWright is a support tool, not a medical device. It identifies that a child’s writing process differs from their own baseline and from their peers, and hands that evidence to a qualified professional who decides what it means. No automated grading, no automated placement, no automated diagnosis — and a teacher always makes the final decision.

For a system, not just a classroom

What changes at policy scale.

the window that is normally lost Outcome-only marking the product “the work looks acceptable” ✕ output visibly fails remediation begins here Process signal reading the production ✓ divergence visible support and referral begin here Year 1 Year 2 Year 3 Year 4 Year 5 Schematic — illustrates the mechanism, not measured timings.
Support that arrives after the output fails is remediation. Support that arrives when the process first diverges is prevention. The distance between those two moments is measured in school years — which is why a process signal is a system-level intervention, not a classroom convenience.

The intervention window moves earlier

Support that arrives when written output has already failed is remediation. Support that arrives when the process first diverges is prevention — and the difference is measured in school years, not weeks.

Teacher time returns to teaching

Lesson material, curriculum-aligned assessment and the class report are generated and pre-checked. The teacher’s judgement is spent on the decision, not on the preparation.

Inclusion becomes the default path

One common activity for the whole class, adapted by need — the legal and pedagogical opposite of a separate track. Accessibility is built into the content model itself, not added as an overlay widget.

Status & governance

What is running, what is being built, and the lines we do not cross.

Stated plainly, because a reviewer should not have to work it out. Nothing on this page is described as shipped unless it is.

CapabilityProductStatusDetail
AI learning board — curriculum-locked lessonsAprendiz● in serviceBrazilian public schools · years 5, 7, 8 science
AI quiz generation — state-assessment item format, Bloom-taggedAprendiz● in servicePaper (OMR) or on screen; every item curriculum-checked and teacher-reviewed
Experiment & investigation tabsAprendiz● in serviceHypothesis · manipulated variable · argued conclusion; model limits disclosed
AI-rendered science images and videoAprendiz◆ in service, expandingGenerated from approved material, checked against the lesson text, teacher sign-off
Interactive simulations (PhET foundation)Aprendiz◐ in development5 activity types classified; 3 live covering 46 of 61 activities; content awaiting specialist validation
Accessibility — 4 adjustable axes + non-negotiable baselineAprendiz◆ rolling outWCAG 2.2 AA design target · eMAG · LBI 13.146 · LGPD · LIBRAS · BNCC · UDL
Subject expansion — arts, geography / historyAprendiz◐ in developmentBuilt on national open data and public-domain collections
Pen capture & writing-process replayVeriWright◆ pilot-readyDigital pen on ordinary paper; demonstrations on this hub use synthetic data
Cognitive-process dashboard for teachersVeriWright◐ controlled POCPlanning / translating / monitoring / revising, with cost per session
Class report, three-band tiering & referral packVeriWright◐ controlled POCValidates operability, teacher workflow and data safety before scale
Authorship / integrity signalVeriWright○ research-stageAuthorship continuity, not AI-detection; no automated consequence
Inclusion programme — individual learning plan, planning & trainingBoth axes◐ POC in KoreaPlan → practice → re-measurement; in controlled POC with Korean partners, preparing launch

The lines we do not cross

  • The teacher decides. No automated grading, discipline, placement or diagnosis, in either product.
  • Screening is not diagnosis. A flag routes a child to a qualified professional; it never labels them.
  • One class, one activity. Adaptation by need, never a separate track — the LBI 13.146 principle, applied in the content model.
  • Compared with themselves. Process indicators are read against the same child’s history, not ranked between children.
  • Data minimisation. Minimal collection, limited retention, deletion by default; no re-use for unrelated model training, advertising or sale.
  • Rights are real. Students and families can request explanation, correction, appeal and human review.
  • Curriculum-locked generation. Content is generated from approved material only, and reviewed by a teacher and a subject specialist before a student sees it.
  • Every demonstration is synthetic. No real child’s record appears anywhere on this hub.

Inclusion programme · POC in Korea · preparing launch

A measurement is only worth taking if it becomes a plan.

Reading the process tells a school what is happening. It does not, on its own, tell a teacher what to run on Tuesday morning. The third piece of the framework closes that gap for the students inclusion policy is written for — an individual learning plan, a planned week, and a training programme, generated from the same signals the two axes already produce, and re-measured against the same baseline. It is in controlled POC in Korea and being prepared for launch.

01 · Read

Signals already on file

The science profile and the writing process for that child — plus what the school already knows. No new testing round is imposed to start.

02 · Plan

One individual learning plan

Goals in the child’s own terms, the supports switched on by need, and what “progress” will look like — in one document a teacher, a specialist and a family can all read.

03 · Schedule

A planned week, not a wish list

The plan is broken into what fits an actual timetable: which activity, in which lesson, with which adaptation — inside the class the child is already in.

04 · Train

Targeted practice

Short, repeatable training on the specific thing the process showed — sequencing, planning before writing, sustaining attention, self-checking — for school and, where a family wants it, for home.

05 · Re-measure

Against the same baseline

The same indicators, read again after the cycle. The plan is updated by evidence, not by impression — and a plan that did not work is visible as such.

One programme, not three purchases. Science, writing and inclusion support run on the same profile, the same accessibility contract and the same reporting surface — so a school adopts one thing and a system procures one thing. This is what makes the framework an integrated learning programme rather than two products that happen to share a logo.

Status, stated plainly. The inclusion programme is in controlled POC in Korea with partner institutions and is being prepared for launch. It is not in service in Brazil or Paraguay, and nothing on this page should be read as a shipped capability. As everywhere else in the framework: the plan is a proposal to a qualified professional and a teacher — it is never an automated placement, and never a diagnosis.

Go deeper on either axis.

Each product page carries the live demonstration, the worked examples and the full accessibility and governance detail behind this overview.