ScoreFluent

Pass your immigration English exam with confidence and true fluency. The best in class AI-powered coach that guides you at every step.

Comment

Inspiration

I moved countries on the back of an English exam, so I know firsthand what students carry — the pressure, the money, the stakes. High-quality, private English tutoring is financially out of reach for most students in South Asian and Southeast Asian countries. I watched the test-prep industry milk that financial anxiety with leaked prediction questions, memorised templates, and cheap hacks to game the scoring. I am ethically against it. It fails students twice: once at the test centre when the tricks don't hold, and again when they arrive in a new country without the language fluency they actually need.

What made it personal: after migrating to Australia, someone close to me — despite having passed her exam — admitted she lacked the confidence to speak English day-to-day. That gap between test-passing and true fluency is our whole thesis. It has only gotten more urgent: Pearson's PTE scoring engine updated twice recently (Nov 2024 and Aug 2025) specifically to penalise the templated answers that entrenched prep platforms still teach.

What it does

ScoreFluent is a cross-platform (iOS, Android, and web) AI coach for PTE Academic, TOEFL iBT, and LanguageCert Academic. It evaluates Speaking, Writing, Reading, and Listening, with each exam scored against its own official rubric.

  • Practise free, pay for depth: Instant baseline scores are always free. Non-expiring credits unlock comprehensive, examiner-grade reports with no subscription traps.
  • Best-in-class AI models: Utilises cutting-edge language models to power a personalized interactive tutor and deliver hyper-specific feedback.
  • Actionable, targeted improvement: No generic "7.5, keep practising" responses. The system pinpoints specific weak areas and generates custom practice questions tailored to strengthen those exact gaps.
  • Honest score prediction: Accounts rigorously for model uncertainty, refusing to display a predicted score number below the evidence floor.
  • B2B Architecture: A white-label layer for educational institutes built with strict, server-enforced role separation.

How we built it

The application uses Expo, Express/tRPC, Postgres (Neon/Prisma), and BullMQ/Redis, alongside a custom Python speech pipeline hosted on Modal. Claude handles writing and speaking grading and powers the interactive tutor. Gemini Flash on Vertex AI dynamically generates the practice question bank in production, acting as the core inventory supply chain.

To automate operations, I deployed an autonomous agent fleet on Cloud Run powered by Gemini. This fleet handles:

  • Resilient Monitoring: Detects unexpected platform outages in real time, automatically emailing affected users with an apology and free system credits.
  • Inbox Automation: Manages marketing and support email inboxes end-to-end.
  • Growth Analytics: Periodically tunes the growth funnel by analyzing user drop-off points and identifying high-traction practice topics.
  • Natural Language Operations: Exposes an NL analytics interface, allowing me to query metrics directly from Slack (e.g., "How many users were active on our platform today?") and receive immediate answers.

Challenges we ran into

  • Not being a credentialed English teacher makes most institutes skeptical.
  • PTE has abundant free content — the fight is for attention, not proof of quality.
  • Competitors source paid prediction questions from student communities I've chosen not to compete on.
  • A growth funnel I misdiagnosed as a pricing problem when it was an unmeasured day-2 retention problem.

Accomplishments that we're proud of

  • Three exams shipped solo; Gemini live in production generating real inventory; two-tier scoring with auto-refund; honest, evidence-sized confidence intervals; a first real B2B trial in progress; and catching my own instrumentation blind spots before they cost revenue.
  • Have made ~30 sales pitches, none have converted yet but 4 are ongoing and in talks.
  • Automated agent driven growth & cold reach pipelines. 200+ user signups, a seasoned ESL teacher and influencer with 100k+ followers is excited about potential partnership, in talks with major exam bodies to use it as a partner preparation tool.

What we learned

  • Students heavily focus on speaking and writing. Those are the areas which make them nervous and they really value actionable improvement steps on it.
  • Educational space relies heavily on organic growth. Its difficult to sell without proper credentials. Trust-building data can't be backfilled log it before you need it, whether that's scoring calibration or a growth funnel.
  • PTE's main audience is in India, Nepal and Sri Lanka. These countries are high volume but very price sensitive.
  • Facebook ad's won't land serious and paid converting audience.
  • We need continuous feedback from heavy users of this space.

What's next for ScoreFluent

  • Growth -
  • Want to bring an ESL teacher/Advisor onboard to help evolve the product, and put the product out.
  • Convert atleast one B2B institute to use this as the primary product.
  • Be an official practice app for one of the ESL exam

  • Collect evidence of success from students and have organic growth on social media via helpful content.

  • Product - We want to bootstrap other ESL exams like IELTS, MET, CELPIP etc. Then train our models to ship a Placement test, which is a mix of these, lasts maybe 1 hr and tells the student which exam is best suited for their strengths and weakness.

  • Engineering - The speaking pipelines have a lot of room for improvement in terms of latency and feedback. Speaking evaluation model needs to be fine tuned for different exam scoring mechanisms. Need a better golden eval set to compare against various models.

Built With

Share this project:

Updates