Tech Blogs Digest 24.11 - 30.11
This week we AI-analysed 7465 posts for you, filtered out the chaff and hand-picked the wheat. Subscribe to stay up to date with future digests!
This week
💻 AI-assisted coding - Claude Code checkpoints, a self hosted PR reviewer, and a Commodore 64 AI game challenge
🏗️ Architecture - Chaos resilient systems
🤖 LLMs in production - A data format to save tokens, an in house content factory, and the nightmare of building an “agentic” AI
📊 Data - The limitless capabilities of Apache Cassandra, and surgically precise updates
🛡️ Security - Building a credential leak monitoring platform, and learning the math essentials for security
🧠 ML - Solving a “simple” NYT puzzle
⚙️ DevOps - Running thousands of iOS pipelines every day
🏗️ ARCHITECTURE
What if the only way to ensure your system survives chaos is to “break it on purpose”? This piece traces how deliberate fault injection, from NASA’s early experiments to Chaos Monkey and modern chaos-engineering tools, became the backbone of resilient, failure-tolerant infrastructure
💻 AI-ASSISTED CODING
Claude Code /rewind: 5 Patterns After a 3-Hour Disaster | 10 min read
When a 3-hour coding mess threatened to cost weeks of work, the “/rewind” feature in Claude Code didn’t just undo - it offered surgical recovery by letting the author rewind either code, conversation context, or both, based on five key patterns that can turn total disaster into a minor speed bump
Every GitLab merge-request now gets an instant, in-house AI review - thanks to a locally hosted Qwen2.5-Coder running inside CI, the team eliminated API fees, kept code private, and still got fast feedback
AI just shrank decades, seeing Google Gemini 3 build a fully playable Snake in 6510 Assembly for the Commodore 64 - combining hot-reload workflows, memory-level control, and real-time emulation
🤖 LLMS IN PRODUCTION
A compact new format called TOON slashes token-use by up to 60% while still preserving full JSON semantics - ideal for feeding structured data to LLMs with less cost and clutter
RhinoGPT : An Experiment in Bringing LLMs to CAD | 16 min read
A free plugin called Rhin oGPT uses LLMs to auto-generate Python scripts that build 3D models in Rhinoceros 3D - surprisingly strong at procedural forms but still a few prompts away from reliable furniture- or house-level design
Building agentic AI sounds fun until bugs, hallucinations and brittle tools turn a weekend project into chaos
A “HIPAA-safe” RAG system still grounded in semantic search - this article shows how de-identifying clinical notes, enforcing access control at retrieval, and re-identifying only for authorized users can prevent data leaks while enabling powerful AI queries on patient histories
A solo creator built an “automated content factory” in their bedroom - churning out blog posts, thumbnails and social-media posts with minimal effort
A careful evaluation of real-world dangers when using LLMs for things like scam detection shows how simple accuracy scores hide big flaws - the article reveals why context, transparency, and robust benchmarking matter more than headline model ratings
Building a Model Context Protocol Server for Hasura: Bringing Natural Language to GraphQL Management | 8 min read
A “model-context protocol server” makes it possible to ask GraphQL APIs in plain English - this article explains how to plug natural-language queries directly into Hasura for seamless, human-friendly data access
A Playground for testing Voice AI Agents | 21 min read
A new playground lets you test voice-AI agents in real time - seeing how speech, context, and AI logic interact before deployment
📊 DATA
Built from scratch at Facebook to handle massive data loads and never go down, Apache Cassandra proved so scalable and resilient that even giants like Netflix and Apple trust it - learn how its masterless, peer-to-peer design smashed limits others couldn’t
StarRocks Incremental MV: A Bridge Over Shifting Ice | 12 min read
By updating only what actually changes, not entire tables, the new incremental-MV framework in StarRocks promises dramatically faster, more efficient data views, making real-time analytics feel within reach (even before full support is released)
🛡️ SECURITY
How “real-time” dark-web credential monitoring often just delivers overpriced “you got breached” alerts after attackers already moved on - and why identity controls beat panic emails
The Security Researcher’s Guide to Mathematics | 26 min read
Math once felt pointless - until I saw how deep algebra, linear algebra, and abstract math power real-world security tools (from machine learning to zero-knowledge proofs)
🧠 ML
What feels like a “light” logic puzzle quickly devolves into chaos - the author built a full-scale solver after realising playing NYT Pips meant 40-minute death matches with colorful dominoes and snaking regions
⚙️ DEVOPS
Building Mac Farm: Running 2000+ iOS Pipelines Daily | 12 min read
At Trendyol, engineers built a custom macOS “farm” of 130 Macs running 2,000+ iOS pipelines daily, executing 50,000+ unit tests and 10,000+ UI tests, to scale development far beyond what traditional CI tools allowed