The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Back to feedPost
just shipped the core of Skillship — an AI-powered LMS built for schools.
One platform, many schools, and one rule that shaped every decision: School A must never see School B's data. Tenant-scoped models, filtered querysets at the base class, UUIDs everywhere.
Four AI features live:
— Career Pilot: personalised career paths for students
— Question generator: chapter PDF in, board-aligned MCQs out
— Adaptive quizzes: difficulty follows actual performance
— Semantic search: teachers ask in plain English, pgvector finds the material
Django + DRF for the LMS core, a separate FastAPI service for all AI work (Gemini + a pgvector RAG pipeline), Next.js 14 on the front. Django brokers every AI call — the AI layer never touches the browser.
The hard part wasn't the AI. It was making multi-tenancy boring enough that no future dev can leak a school's data by accident.
Open for new work — full-stack builds, AI integrations, RAG pipelines, multi-tenant SaaS. Send me a project.
The tenant boundary is where I'd put the first failure test. I'd try to make a user from one school retrieve another school's record before I trusted any of the AI features.
Yes, we already have a multi-school setup in place. The platform currently has 11 schools, 8,200+ students, onboarded/managed through the system. We have also already tested the tenant boundary to ensure that users from one school cannot access records belonging to another...
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Challenges
View all
Related posts
How I Actually Build Websites — From Inspiration to Final Build.
I don’t start with a blank screen. I usually start by exploring Pinterest to find visual references, layouts, interactions, and design ideas. Then I bring the reference into ChatGPT, explain what I want to build, and use it to turn the rough idea into a much clearer development prompt.
From there:
Pinterest → Reference → ChatGPT → Refined Prompt → Antigravity → Manual Refinement → Final Website
I give Antigravity the reference image + refined prompt, let it handle the first implementation, and then take over manually to fix the details that AI doesn't get right — spacing, responsiveness, interactions, typography, animations, and overall polish. AI speeds up the process.
But the final quality still comes from the decisions and corrections I make.
This is how I combine design inspiration, AI-assisted development, and manual frontend work to turn an idea into a real website.
I like the deliberate handoff between reference, refined prompt, and manual refinement—especially the note that the final quality comes from your corrections. That balance is what keeps AI-assisted builds feeling intentional.
As a Frontend Engineer, I treat the product’s invisible context as part of the interface. For Orbyt, I worked on the application layer around a curated healthcare network, including the waitlist, membership flow, and the context connecting operators, doctors, and opportunities. I keep integrations explicit and use read-back to verify what the system actually returned. If your product needs a frontend people can trust, message me.
Amazing work here....The usage of color and elements are sync works well.
Just shipped BladeReveal 🗡
A cinematic slider for Framer with a custom cursor click left or right, the cursor flips direction, the slide sweeps in through staggered blades with Ken Burns zoom.
Navigation you can actually see, not a hidden click zone.
Live now 👇
Visible navigation cues are a good complement to cinematic transitions.