The Problem Nobody Talks About in EdTech
Professional certification training centers — the kind that prepare engineers for PMP, UPDA, PMI, and other Gulf-region certification exams — are drowning in tools. They use one platform to manage courses and materials, another for mock tests, a third for student communication, and a patchwork of spreadsheets for CRM and invoicing.
Every tool has its own login, its own database, its own pricing model. Student data is scattered. Trainers can't see who's failing their mock tests before it's too late. Finance teams manually reconcile payments in Excel. Training Consultants who bring in students on commission have no visibility into what they've earned.
“The problem wasn't finding an LMS. The problem was that no single platform connected the student journey from first inquiry to certification — and the ones that came close charged per-student fees that made scaling prohibitively expensive.”
This is the problem we set out to solve. Not just as software developers — but as a product company building infrastructure that training centers in Qatar, the UAE, and the broader Gulf region can own outright, for a fixed annual fee, with their own branding.
What We Built
The platform we delivered is a complete, production-grade EdTech system. Not a demo. Not a prototype. A live system handling real students, real payments, and real exam preparation:
🎓 Student Portal
Course materials, mock test access, results history, offline PWA support
⚙️ Admin Panel
Student management, enrollments, batch scheduling, full finance overview
👨🏫 Trainer Portal
Course content, session attendance, Zoom integration, schedule view
🤝 TC Portal
Training Consultant leads, commission tracking, enrollment KPIs
📝 Mock Test Engine
Redis-backed live sessions, offline IndexedDB, image-based questions, auto-submit
📊 CRM
Lead pipeline, configurable workflow, activity timeline, lead-to-enrollment conversion
💰 Finance Module
Multi-installment payments, receipt + invoice PDFs, expense tracking, KPI dashboard
📱 Mobile App
Flutter (iOS + Android), same mock test experience, offline test support
The backend is a NestJS 11 API with Prisma 7 ORM, 28+ database models, full RBAC across six user roles, Redis for live test sessions, and AWS S3 for file storage. The frontend is Next.js 16 with PWA support. Everything runs on Docker Compose on AWS EC2, with automated backups, AMI snapshots, and CloudWatch alerting.
The Part That Changed Everything: Claude Code
Here's the honest answer to how a one-developer team shipped this in 12 weeks: Claude Code by Anthropic.
Not as a code autocomplete tool. Not as a search-engine replacement. As a genuine engineering collaborator that could hold the entire context of a complex system in its working memory and make decisions at the architectural level.
When we described a multi-role permission system where ADMIN can see all leads, SALES_EXECUTIVE only sees their own, and SUPER_ADMIN spans all tenants — Claude Code wrote the NestJS guard, the Prisma query filter, the JWT payload shape, and flagged a subtle edge case we hadn't considered, all in a single response.
What We Actually Used It For
- Schema design:Described business rules in plain English; received fully normalised Prisma models with correct relations, indexes, and migration SQL
- Complex business logic:Commission calculation, installment ledger balancing, mock test scoring with per-question weighting — all described once, implemented correctly
- Bug diagnosis:Pasted a 4-line error trace, received the exact root cause and a one-line fix. Consistently.
- API design:"Add an endpoint for Training Consultants to see only their commission for this month" → complete controller, service, DTO, and guard written instantly
- DevOps:AWS IAM policies, SNS alerting scripts, EC2 DR runbooks, Docker Compose multi-service health checks — written and explained in minutes
- Refactoring:Large-scale changes (adding a Branch model across 6 modules) planned, executed, and verified without introducing regressions
“The experience closest to using Claude Code is pair programming with a senior engineer who has already read every line of your codebase and never needs a break.”
The Build Timeline
Weeks 1–2
Foundation
NestJS monorepo, Prisma schema (28 models), auth (OTP + password), multi-tenant architecture, RBAC, JWT strategy
Weeks 3–4
Core Portals
Admin panel, student portal, course management, batch scheduling, Next.js PWA setup
Weeks 5–6
Mock Test Engine
Redis sessions, question bank, DOCX parser (FastAPI), image extraction to S3, offline IndexedDB, scoring algorithm
Weeks 7–8
CRM + Finance
Lead pipeline, configurable workflow, enrollment + payment flows, PDF generation, commission tracking
Weeks 9–10
Trainer + TC Portals
Trainer attendance, Zoom integration, TC KPI dashboard, session management, branch support
Weeks 11–12
Mobile + Production
Flutter mobile app (iOS + Android), Docker deployment, AWS monitoring, go-live
Traditional Timeline vs. AI-Assisted
| Scope | Traditional Team (5 devs) | With Claude Code (1 dev) |
|---|---|---|
| Schema design + migrations | 2–3 weeks | 2 days |
| 8 REST API modules | 8–10 weeks | 3–4 weeks |
| 4 frontend portals | 10–12 weeks | 4–5 weeks |
| Mock test engine (Redis + offline) | 3–4 weeks | 1 week |
| Flutter mobile app | 6–8 weeks | 2 weeks |
| DevOps (Docker, backups, monitoring) | 2–3 weeks | 3 days |
| Total (estimated) | 9–12 months | 12 weeks |
Key Insight
The bottleneck in software development has never been typing speed — it's the cognitive overhead of holding a complex system in your head while writing new code. Claude Code removes that bottleneck. The developer's job shifts from “writing code” to “making decisions and verifying outcomes.”
What This Means for EdTech Startups
We're not claiming Claude Code writes production software without human judgment. It doesn't. Every architectural decision, every security review, every business logic verification — that's still human work. The developer has to know what they're building and why.
What Claude Code changes is the cost of implementation. When implementation is fast and reliable, the constraint becomes product thinking. And that's exactly where human expertise should be focused.
For EdTech startups specifically: the days of choosing between an expensive off-the-shelf platform (with its limitations and per-seat pricing) and a 9-month custom build (with its risk and cost) are over. A small, domain-expert team with Claude Code can now deliver the custom platform in the same timeframe as an enterprise procurement cycle.
Our platform now serves a professional certification training center in the Gulf region, replacing three separate paid tools with a single white-labeled system they own. They pay a fixed annual fee. No per-student costs. No vendor lock-in on their student data. As they grow from 500 students to 5,000, their software cost doesn't change.
What's Next
We're now offering this platform — built on the same codebase, fully customisable — to other professional training centers across the Gulf and South Asia. The white-label model means each customer gets their own branding, their own domain, and full ownership of their data. If you're running a training center and you're tired of managing five different tools — we should talk.
See What We Built — Or Let's Build Yours
Whether you want to see a live demo of EduOS or explore how Saravonix can build something similar for your business — we'd love to connect.
Karthik Balasundaram
Founder & CTO, Saravonix
Building product infrastructure for the Gulf EdTech and professional services market. Passionate about using AI to make world-class software development accessible to small, expert teams.