GitHub Weekly: Architecture & Infra Tools Worth a Look (September 14 – September 20, 2026)

TL;DR: This week we’re setting aside the usual overall-trending list to spotlight repos tagged around architecture, system design, diagramming, and infrastructure-as-code — the stuff most relevant to readers thinking about how systems are put together. The standout theme: a cluster of new “diagram-as-code” tools built specifically as skills for AI coding agents like Claude Code and Codex, alongside a few solid system-design learning resources.

Instead of chasing whatever’s viral across all of GitHub, this week’s digest is scoped to repos tagged architecture, system-design, diagram, and infrastructure-as-code — a deliberate match for a blog that spends most of its time on IT architecture. Covering September 14 through September 20, 2026, here are the 10 repos that came out on top.

#1. coldteadotai/pr-lens

github.com/coldteadotai/pr-lens — TypeScript, 1242 stars

Lens takes a pull request and automatically draws it as an animated architecture and data-flow diagram, right inside the PR itself. Instead of reading a wall of diffed code and mentally reconstructing what changed and how it flows through the system, you get a visual walkthrough generated for you. It ships as a GitHub App, a GitHub Action, a CLI tool, and — notably — as a “skill” that AI coding agents (like Claude Code or Codex) can call directly.

Good for: Teams doing code review on architecturally significant changes, and anyone who finds it faster to look at a diagram than to trace call chains across a dozen changed files.

#2. xevrion/breakscale

github.com/xevrion/breakscale — TypeScript, 1169 stars

Breakscale is a browser-based system design simulator: you build out a distributed system on a canvas — load balancers, databases, caches, queues — then throw traffic at it and watch where it actually breaks. It’s built on real queueing theory under the hood, so the failures you see (a database falling over, a queue backing up) reflect genuine bottleneck behavior rather than a scripted demo.

Good for: Engineers studying for system design interviews, and teams who want a hands-on way to teach junior developers why certain architectures buckle under load before it happens in production.

#3. DrHazemAli/enterprise-system-design

github.com/DrHazemAli/enterprise-system-design — No listed language, 632 stars

This is a large, source-grounded reference course covering what it actually takes to design systems that survive real-world conditions: partial failures, security review, changing requirements, and traffic spikes. It spans distributed systems, cloud and edge architecture, AI systems, cybersecurity, and reliability engineering, and it cites real standards and frameworks rather than just offering opinions.

Good for: Senior engineers and architects who want a structured, well-researched refresher on enterprise-grade system design, or anyone mentoring others through that material.

#4. bybit-exchange/svg-diagram

github.com/bybit-exchange/svg-diagram — JavaScript, 577 stars

Another entry in this week’s “diagram tools built for AI agents” theme: svg-diagram is an agent skill that draws architecture, flowchart, sequence, and data-flow diagrams as hand-placed SVG following one consistent, linted house style. Rather than relying on auto-layout tools like Mermaid that can produce cluttered output, it aims for diagrams that look deliberately composed.

Good for: Teams that want their AI coding assistant to produce clean, consistent architecture diagrams for documentation without a human redrawing everything afterward.

#5. quickdrawjs/quickdraw

github.com/quickdrawjs/quickdraw — JavaScript, 479 stars

Quickdraw is an MIT-licensed infinite-canvas whiteboard SDK — essentially the building blocks behind tools like tldraw or Excalidraw, but free to embed in commercial products with no license fee. It ships bindings for React, React Native, and plain JavaScript, so you can drop a full drawing/diagramming canvas into your own app.

Good for: Product teams building internal tools (design canvases, architecture whiteboards, collaborative diagramming features) who don’t want to pay licensing fees or build a canvas engine from scratch.

#6. lazypay/Archscribe

github.com/lazypay/Archscribe — Python, 351 stars

Archscribe generates polished, hand-drawn-style architecture and process diagrams with a dark background, and it’s built as a skill for coding agents like Codex and Claude. It outputs an editable Excalidraw file alongside a PNG and an actually-animated GIF, so the diagram isn’t just a static picture — it can show data or requests moving through the system.

Good for: Anyone who needs presentation-ready architecture diagrams for a slide deck or blog post (like this one) without manually laying out boxes and arrows.

#7. design-gurus/grokking-system-design

github.com/design-gurus/grokking-system-design — No listed language, 288 stars

This is the free, open companion to DesignGurus.io’s well-known “Grokking the System Design Interview” course. It walks through the standard system design interview patterns — designing a URL shortener, a rate limiter, a chat system, and similar classics — in a structured, free format.

Good for: Developers preparing for system design interviews at larger companies who want a free, community-maintained alternative to paid interview-prep courses.

#8. amitshekhariitbhu/transformers-explained

github.com/amitshekhariitbhu/transformers-explained — No listed language, 217 stars

A step-by-step breakdown of the Transformer architecture — the model design behind most modern large language models. It covers the full architecture, every major attention variant, positional embeddings, and what each layer is actually doing, aimed at building real intuition rather than just reciting the “Attention Is All You Need” paper.

Good for: Developers who use LLMs daily but have never dug into how they’re actually built, and want to understand the architecture well enough to reason about model behavior and tradeoffs.

#9. ombharatiya/AI-Engineer-Interview-Questions

github.com/ombharatiya/AI-Engineer-Interview-Questions — Python, 130 stars

A structured collection of interview questions and prep material for AI Engineer, LLM Engineer, and Applied AI roles. It includes company-specific question sets, role-based study maps, worked-through answers, and from-scratch coding challenges covering things like RAG (retrieval-augmented generation, where a model looks up external data before answering) and prompt engineering.

Good for: Backend and software engineers trying to break into AI-focused roles, and ML practitioners who want to level up toward more applied, production-facing AI engineering positions.

#10. MengTo/towers

github.com/MengTo/towers — HTML, 125 stars

A change of pace to close out the list: this is a single self-contained HTML file that procedurally generates and animates a tower — a Japanese tenshu, a Chinese pagoda, or a Thai prang — building itself from the ground up in about four and a half seconds using WebGL. It’s actual architecture in the literal, building-design sense rather than software architecture, and it’s a fun example of generative, rule-based design packed into one file with no dependencies.

Good for: Developers curious about procedural generation and WebGL, or anyone who wants a bite-sized, dependency-free example of building a 3D scene from code.

The clearest pattern this week is the rise of diagram-generation tools built specifically as skills for AI coding agents (pr-lens, svg-diagram, and Archscribe all fit this mold) — a sign that “explain this codebase visually” is becoming a first-class task for agents, not just humans. Pair that with a few solid system-design and LLM-architecture learning resources, and it’s a good week to bookmark something even if you don’t have time to dig in today. See you next Monday for another angle on what’s trending.