The Challenge
Hong Kong's private tutoring market is worth billions, yet finding the right tutor remains frustratingly manual. Parents rely on word-of-mouth, WhatsApp groups, and community notice boards. Tutors, especially qualified freelancers, have no efficient way to reach students beyond their personal networks.
The founder of HK TutorLink came to Astera with a clear vision: their existing website was severely outdated — a legacy site with no mobile responsiveness, poor UX, and purely manual tutor matching. They needed a complete rebuild: a modern platform where students could discover verified tutors, compare qualifications and reviews, book sessions online, and pay securely — all in one place. The goal? Transform the outdated site into a competitive, AI-powered platform before the start of the academic year.
Our Approach
Phase 1: Discovery & Architecture (Weeks 1–2)
We started with a 5-day discovery sprint: stakeholder interviews, competitor analysis (covering both HK-local platforms and international models like Wyzant and Superprof), and user journey mapping. The output was a prioritized feature backlog, database schema, and system architecture — all approved before any code was written.
Key architectural decisions made early:
- React + Next.js for the frontend — SSR for SEO (tutor profiles need to rank on Google), fast navigation, bilingual routing
- Node.js + Express backend with PostgreSQL — relational data model fits perfectly for tutors, students, bookings, and reviews
- Stripe Connect for marketplace payments — tutors get paid directly, platform takes a commission, full PCI compliance without building payment infrastructure
- AWS (EC2, RDS, S3, CloudFront) — scalable, cost-effective, data stays in the APAC region
Phase 2: Design & Prototyping (Weeks 2–3)
We designed the full user experience in Figma — tutor onboarding flow, student search and discovery, booking calendar, payment flow, and review system. Every screen was designed in both English and Traditional Chinese from the start, not retrofitted. The founder tested the interactive prototype with 10 real parents and tutors before development began.
Phase 3: Build & Iterate (Weeks 3–9)
Six weeks of agile sprints with weekly demos to the founder. Every Friday, a new set of working features was deployed to staging:
- Sprint 1–2: Tutor registration, profile creation, verification workflow, admin dashboard
- Sprint 3–4: Student search with filters (subject, district, price, language), AI-powered matching algorithm, tutor profile pages with SEO optimization
- Sprint 5–6: Booking system with calendar integration, Stripe Connect payments, review & rating system, email/SMS notifications
The AI matching component uses a recommendation engine that weighs multiple factors: subject expertise, location proximity, price range, availability overlap, student learning style preferences, and historical success rates. This is powered by a custom algorithm built on PostgreSQL's pgvector extension, not a generic LLM — because matching quality is the platform's competitive advantage.
Phase 4: Launch & Growth (Week 10+)
We handled the full launch: production deployment on AWS, monitoring setup (Sentry for error tracking, Grafana dashboards for key metrics), SEO optimization for tutor profiles (each profile is a unique, indexable page with structured data), and App Store submission for the progressive web app. Post-launch, we continued on a monthly retainer for feature additions, performance optimization, and growth experiments.
Tech Stack
- Frontend: React, Next.js, TypeScript, Tailwind CSS
- Backend: Node.js, Express, PostgreSQL, Redis
- AI/ML: pgvector, custom recommendation engine
- Payments: Stripe Connect (marketplace model)
- Infrastructure: AWS (EC2, RDS, S3, CloudFront), GitHub Actions CI/CD
- Monitoring: Sentry, Grafana, AWS CloudWatch
The Results
Within 3 months of launch:
- 500+ active users — students and tutors actively using the platform
- 200+ verified tutor profiles — each with qualifications checked and background verified
- AI matching accuracy of 85%+ — measured by booking conversion rate from recommendations
- Zero downtime since launch — monitoring and auto-scaling ensure 99.9% uptime
- Organic search traffic growing 40% month-over-month — thanks to SEO-optimized tutor profiles ranking for long-tail queries like "physics tutor Kowloon"
The platform is live at hktutorlink.hk and continues to grow with ongoing development from the Astera team.
Key Takeaways
- Start with discovery, not code. The 5-day discovery sprint saved weeks of rework by aligning everyone on scope before development began.
- Design bilingual from day one. Retrofitting Chinese support after the fact is 3x more expensive than building it in from the start.
- AI doesn't have to mean LLMs. The right AI for this problem was a recommendation engine, not a chatbot. Pick the tool that fits the problem.
- Weekly demos keep everyone aligned. The founder never felt out of the loop, and early feedback caught UX issues before they became expensive to fix.
"Astera took my idea and turned it into a real product in 10 weeks. As a non-technical founder, I was worried about being lost in jargon — but every weekly demo was in plain language, and I always understood exactly where we were. The AI matching is the reason tutors and students keep coming back."
Have a Similar Project?
Whether you're building a marketplace, a SaaS platform, or an internal tool — we'd love to hear about it. Book a free consultation to discuss your project scope, timeline, and budget.