We build the engines that testing and training companies run on — computer-adaptive testing, real-blueprint mock exams, and adaptive practice that’s honest about small-sample reality.
Four areas where our work goes deeper than a portfolio screenshot can show.
The engine picks the next question based on what the candidate just answered. Done well, it shortens the exam without losing accuracy — and the routing logic itself is something we can explain in plain English.
Proof — Practice & mock exam platform for AMC candidates in Australia, with a CAT engine choosing the next question.
Most practice sessions are 5–50 questions — too few for the heavy statistical models most platforms claim. We use threshold-based methods that answer "you're at 62% on air brakes" in plain English.
Proof — QuizSprint and PassHHA both run on this.
If practice doesn't mirror the actual exam's category mix, candidates walk in surprised. Our PassHHA mock exam matches the HHA blueprint (Personal Care 24%, Safety 20%, …) exactly.
Proof — PassHHA.
Low-vision support, simplified interaction patterns, alternative answer flows. Practice systems usually optimise for the median candidate; we've built for the ones the median ignores.
These aren’t variations of one adaptive engine we slap on every client. They’re three deliberately different systems — because an HHA candidate, a Malaysian SPM student, and a CDL career-changer each need the test to behave differently.
The pattern is the lack of a pattern. Scroll for the three worlds, side by side.
A solution has pH 4. How does [H+] change if pH becomes 2?
When assisting a client with bathing, the aide notices a reddened area on the lower back that does not fade. What is the most appropriate next step?
Side by side, the three systems share none of the parts. Display family, palette weight, button shape, error-state tone — every choice was made for the candidate, not the template.
We build across the entire learning ecosystem — from the learner’s phone to your team’s dashboard. No generic websites, just modern systems that make education operations scale.
Mock exams that mirror the official blueprint to the percentage point — not invented taxonomies. Readiness scoring with a deliberate safety margin above the real pass line. Weak-topic identification users can interrogate, not a black-box "AI thinks you should study this."
Confidence calibration tracked silently (correctness × self-reported certainty) — a real metacognitive signal. Short-session pacing because adult candidates study in 4-minute bursts on a phone between shifts. Optional time pressure when the exam is timed; off by default when it isn't.
One place for content, progress, billing, certificates, and communication — replacing the spreadsheet-plus-three-SaaS-tools setup most providers run on today. Designed for the operator's day, not just the learner's.
AI where it earns its place: question generation against your blueprint, explanation drafting for new content, large-bank categorization that would take a human team weeks. Not "AI" as a marketing badge bolted onto a chat widget.
Cohort tracking, intervention alerts when a learner stalls, completion reporting that survives an audit. The dashboard your operations team actually opens daily — not the vanity one they screenshot for board meetings.
Mobile-first means designed for slow 4G on a lunch break, not "responsive after we built the desktop version." Fast initial load, offline-tolerant where it matters, large tap targets, no infinite-scroll attention traps.
Enrollment → scheduling → payment → certificate issuance → audit trail. The unsexy operational layer that quietly determines your margin. We design assuming someone will be on leave when an edge case fires.
We don’t only build for clients. We launch and operate our own education products — so the engineering judgments behind them aren’t agency theory. They’re decisions we made under real constraints and live with daily.
Below are four of those judgments, drawn directly from the systems we run.
When a learner asks "why this question?", we want a real answer. We choose threshold-based algorithms over Bayesian Knowledge Tracing when sample size doesn't justify the math — most prep sessions are 5–50 questions, which is noise territory for BKT. A threshold answers in plain English: "you're at 62% on air brakes." A posterior probability cannot.
We don't invent question taxonomies. Our PassHHA mock exam mirrors the official HHA blueprint to the percentage point (Personal Care 24%, Safety 20%, and so on). If our practice reflects the real exam, candidates recognize the shape on test day. Anything else is theatre.
A working HHA candidate, a Malaysian SPM student, and a CDL career-changer each have a different kind of pressure, a different readiness signal, and a different sampling strategy. We resist the urge to apply one adaptive engine to all of them — and we can tell you why each choice was made for each user.
Some questions our code can settle: how the adaptive engine samples, what threshold flags a weak topic, where the data sits. Others it can't: what your learners actually do at 11pm before a high-stakes exam. We tell clients which is which. We build the system. We don't replace the discovery.
Tell us what you’re running today, what’s breaking, and what you wish it did. We’ll come back within two business days with an honest read on whether we’re a fit — and what the work would actually look like.