How to Build an AI-Powered Test Prep Platform

Building an AI-Powered Test Prep Platform: The Phrasel Playbook

When preparing for Canada PR, one truth becomes obvious fast: your Comprehensive Ranking System (CRS) score is everything. Language points make up a large share of that score, which means your English test isn't just another exam — it's a gateway. For many immigrants, it translates directly into CLB levels, Express Entry points, and a timeline for a fresh start.

Then PTE Core launched in 2024. A fully machine-scored exam felt like a better fit for how I think about practice: measurable, consistent, and less dependent on the luck of a human examiner on any given day. The catch? The prep landscape was sparse. Most existing platforms were built for PTE Academic, and they mostly shipped those Academic materials over to Core. The overlap was incomplete, and a lot of it wasn't really relevant for immigration candidates.

I saw that gap as an opportunity. I could build a platform and a community specifically for learners on the Canadian immigration path. Even better, I could put my love for languages and AI into something meaningful for people going through the same journey I was on.

The Problem With Volume-Based Test Prep

The existing prep market is loud on volume:

• Practice 1000 questions
• Memorize these templates
• Watch this 45-minute tip dump

Volume helps until it doesn't. Blind repetition builds confidence that collapses under exam pressure. Templates help until the task doesn't match the template. More material without a diagnosis just makes the haystack bigger. Shipping Academic content into a Core-shaped hole does the same thing: more practice, less relevance.

The Core Thesis: Clarity Over Quantity

My thesis became simple: Learners do not need more practice material. They need clarity on what to improve next.

That line still sits on the About page, because it's the product. Phrasel is focused on PTE Core (not Academic), intentionally. The Canada PR path has different stakes, different score goals, and different workflows than university admissions. So instead of chasing question counts, I focused on feedback loops that turn practice into progress.

What Phrasel Ships Today

Here's what I built:

• Practice across Speaking, Writing, Reading, and Listening: exam-style tasks with guides and examples
• AI scoring for speaking & writing: pronunciation, fluency, content, grammar, vocabulary; word-level feedback where it helps
• Full-length timed mock tests, with a report that shows overall + skill scores, strengths, and focus areas
• Canada PR tooling: CLB conversion, CRS language points, test comparisons, so practice maps to immigration goals
• A free tier: practice + one scored mock, because people should feel the loop before paying

Live at phrasel.com · App at app.phrasel.com

The AI Scoring Engine: How It Works

At the heart of the platform is AI Scoring. For Speaking and Writing, I needed a system that could evaluate performance the way the actual exam does — but faster and with more detail. The AI Scoring stack covers:

• Pronunciation: Phoneme-level analysis using speech recognition models
• Fluency: Measuring pace, pauses, and rhythm against PTE benchmarks
• Content: Topic relevance and completeness scoring
li>Grammar and Vocabulary: Error detection and lexical range assessment
• Word-level feedback: Highlighting specific areas for improvement

This isn't just automated grading. It's diagnostic feedback that tells learners not just what they scored, but why and how to fix it. That's the difference between a number and actionable insight.

From Practice to Progress: The Feedback Loop

I refused to win by having the biggest question bank. The unit of value isn't another unanswered Read Aloud. It's leaving a session with a clear next step. The feedback loop looks like this:

1. Take a task: Choose from Speaking, Writing, Reading, or Listening practice
2. Get scored: AI provides detailed feedback within seconds
3. Review insights: See strengths, weaknesses, and specific improvement areas
4. Target practice: Focus on the skills that need work
5. Track progress: Monitor improvement over time toward target scores

This loop ensures every session moves the learner closer to their goal, rather than just adding to their question count.

Canada PR Tooling: Connecting Scores to Outcomes

One of the biggest gaps I saw was the disconnect between test scores and immigration outcomes. Learners could get a 72 on PTE Core, but what did that mean for their CRS score? For their Express Entry profile? To bridge this gap, I built Canada PR tooling that includes:

• CLB conversion: Automatically map PTE scores to Canadian Language Benchmark levels
• CRS language points: Show exactly how each point increase impacts overall ranking
• Test comparisons: Compare PTE Core performance against IELTS and CELPIP benchmarks

This turns abstract test scores into concrete immigration strategy. A learner can now see that improving their Speaking score from 78 to 82 will push them from CLB 9 to CLB 10, adding 13 points to their CRS score — and potentially changing their entire application timeline.

The Free Tier Strategy

Building trust in AI Scoring is hard. People need to feel the accuracy before they pay. That's why I made the free tier generous: practice + one scored mock test. This lets users experience the full feedback loop without any commitment. The conversion rate from free to paid has been strong because learners can immediately see the value of detailed, actionable feedback.

The free tier also serves as a community builder. Users who get real value from their first scored mock often become advocates, sharing their progress and referring others. Word-of-mouth has been the primary growth channel.

Monetization: Beyond Just Subscriptions

While the core platform operates on a subscription model, the AI Scoring technology has broader potential. The same engine that evaluates PTE Speaking and Writing could power:

• EdTech platforms: Integrating AI-powered assessment into online language schools
• SaaS tools: Offering AI Scoring as a service to other test prep companies
• Micro-SaaS opportunities: Building specialized tools for specific test sections or skill areas

This multi-pronged approach ensures the platform isn't just another test prep site — it's becoming an infrastructure layer for AI-powered assessment in the EdTech space.

Key Lessons for Builders

If you're thinking about building in this space, here are the lessons that mattered most:

Solve for the specific, not the general

Trying to serve both PTE Academic and PTE Core users would have diluted everything. By focusing exclusively on the Canada PR path, I could tailor every feature to immigration-specific needs. The same principle applies to any EdTech or SaaS product — depth beats breadth.

Make the AI explain itself

Early versions of the AI Scoring engine just gave scores. Users didn't trust them. Adding word-level feedback and clear rubric breakdowns changed everything. When learners understand why they got a certain score, they trust the system and act on the feedback.

Design for the outcome, not the activity

Test prep platforms are full of busy work that feels productive but doesn't advance the learner. Every feature at Phrasel asks: "How does this help someone get closer to their CRS target?" That question killed a lot of good ideas that didn't serve the ultimate goal.

Start with a free loop that proves value

The free tier isn't a teaser — it's a proof point. One scored mock test that delivers detailed feedback is enough to convert serious users. Make sure your free offering is substantial enough to demonstrate real value.

The Path Forward

Looking ahead, the plan is to expand the AI Scoring capabilities and explore partnerships with other EdTech platforms. The technology stack — combining speech recognition, natural language processing, and immigration-specific data — positions the platform well for both direct-to-learner growth and B2B opportunities.

The real win isn't in having the most practice questions. It's in creating a system where every minute of practice is optimized for the outcome that actually matters: the score that opens doors.