I've been designing distributed systems for years. Work, interviews, side projects. And every time, same thing: figure out requirements, sketch the architecture, work through data models, draw sequence diagrams, try to find where the whole thing falls apart at scale. Starting from scratch every single time.
At some point I got tired of it and built an AI architect that does it with me. That turned into SysDesAI.
How it started
I was prepping for system design interviews and kept hitting the same wall. ByteByteGo has great videos but I'm just watching someone else design. Grokking has structured lessons but the designs are already written for me. System Design Primer is solid but it's a wall of text on GitHub. And yeah, you can throw questions at ChatGPT, but there's no real workflow, no diagrams, nothing saves.
What I actually wanted was to practice designing systems myself. Sit down, think through the requirements, make trade-off decisions, produce actual architecture diagrams. Basically what you do in a real interview, but without the pressure of someone judging you in real time.
So I built that.
What the AI architect actually does
You tell it what you want to design (a URL shortener, a ride sharing app, a payment system, whatever) and it works in two modes.
Copilot mode is the interactive one. The AI asks you questions like an interviewer would: what are your requirements? how would you handle data consistency here? what happens when this service goes down? You work through each step together, and there's a chat where you can push back, argue with it, tell it to change things. It actually updates the design based on what you say. More like working through a problem with a colleague than getting a lecture.
Autopilot mode generates the whole design end to end on its own. Requirements, capacity estimates, architecture diagrams, data models, sequence flows, scaling strategy. Useful when you want to study how a complete design comes together without going through every step yourself.
One thing that turned out pretty useful: it generates architecture diagrams based on what you actually design. Not pre-made textbook diagrams. And when you're done, you can export the whole thing as a Markdown spec and drop it straight into your AI coding tool (Cursor, Claude Code, Windsurf, GitHub Copilot). Architecture first, then code.
There's also a video interview mode
This one I added later. It's basically a FAANG-style system design interview with an AI interviewer. You talk to it via voice, sketch on a live whiteboard, and it scores you across 6 categories with feedback. Closest thing to a real interview without actually being in one.
The free stuff (seriously free)
I wanted the learning resources to be free, no catches:
- A full software architecture course: 11 modules, 60+ lessons, from scalability basics to distributed consensus. Quizzes in every lesson, completion certificate, AI chat built into each lesson so you can ask questions about the material.
- 48+ architecture walkthroughs you can explore step by step, ask the AI about, or fork to make your own version.
- A curated news feed from 35+ engineering blogs (Netflix, Uber, Cloudflare, Stripe, AWS). Updates every 6 hours. Each article has a "Design this" button that sends it to the AI architect so you can design the system described in the article yourself.
- Community features: discussions forum, architect profiles, leaderboard, reputation levels.
No signup required for any of that.
Pricing (the short version)
Everything except the AI features is free. For the AI architect, chat, and video interviews: you get 50 free credits to try it out. After that you can buy credit packs (starting at $5) or bring your own API key. BYOK means you connect your Google AI or OpenAI key and use everything unlimited. SysDesAI charges nothing, you just pay whatever the API provider charges. Your keys stay in your browser, never logged or stored on our servers.
The BYOK approach is the way to go if you plan to use it regularly.
How it was built
Here's the slightly ironic part: the whole app was built aggressively with AI coding tools (mostly Claude Code). Took about 12 days. A year or two ago that would have been months of work. The architecture is far from perfect, the goal was to iterate fast and build something useful, not to win any clean-code awards. If it keeps growing I'll probably ask the SysDesAI architect to design a scalable architecture for SysDesAI itself.
Try it out
Still a work in progress. Bugs, rough edges, things that could be better. If you give it a try, I'd love to hear what works, what doesn't, what you'd change.
I'm Igor, full stack dev. Built this because I was tired of how passive system design prep is. Happy to answer questions in the comments.
Top comments (0)