UUbaidullah
  • Work
  • About
  • Writing
  • Now
  • Uses
  • Playground
Let's build something that actually ships.

I take on a small number of product engineering engagements at a time. If you're building something AI-shaped and need someone who can own end-to-end, that's where I'm useful.

Start a conversation
Sitemap
  • Work
  • About
  • Writing
  • Now
  • Uses
  • Playground
  • Contact
Elsewhere
  • GitHub
  • LinkedIn
  • Instagram
  • Email
© 2026 Ubaidullah. Built in Pakistan.
Next.js · React 19 · Tailwind v4 · Vercel
All work
LiveNov 2023 — PresentFull-Stack / Product Engineer

Tututor.ai

AI-native education platform used daily by 17,000+ students and teachers.

Visit Tututor.ai
17,000+
Students & teachers
across schools in Murcia, Spain
90–95%
Lesson-prep time saved
in measured teacher workflows
1
Person
owning AI services, CRM & student UX
0→1
AI features
shipped to production users
The problem

Teachers were spending the majority of their week assembling lesson material, quizzes, and remediation content by hand. Existing 'AI tools' produced generic output that needed so much editing they barely saved time. The school's CRM was a separate product entirely, so admins were duct-taping spreadsheets to track classes, content, and student progress.

The approach

I treated this as one product, not two. A small set of AI primitives — chatbots, quiz generation, grading, conversation analytics — composed into teacher-facing tools that share a single content + class model with the CRM. Heavy AI flows live behind a microservices boundary so the teacher UI stays responsive while the LLM is thinking. WebSockets stream partial responses, and conversations are stored structured (not blobs) so teachers can review and search them.

Architecture

How the system is wired.

The boundaries that mattered: keeping the teacher UI responsive while heavy AI work happens behind a WebSocket + microservice boundary.

Client
Service
AI
Data
External
Hover to focus a service
The outcome

Lesson-prep time dropped 90–95% in measured cases. Several schools in Murcia adopted the platform; thousands of students use it daily. The architecture has held as features compounded — adding the quiz generator and conversation analytics didn't require rewriting the core.

What I owned
  • 01Designed the AI chatbot system — teachers spin up topic-specific tutors and review every student conversation to surface gaps.
  • 02Built the AI quiz generator with automatic grading and per-class performance analytics from lesson content.
  • 03Architected the school CRM (students, classes, content) so the AI tools and admin tools share one model.
  • 04Migrated heavy AI flows behind WebSockets + a microservices boundary so teacher UX never blocks on the LLM.
  • 05Acted as UX designer for every teacher-facing flow — no separate designer on the team.
Stack
ReactNode.jsExpressMongoDBWebSocketsMicroservicesOpenAI APIElevenLabs
Next
Nobuko Japan