// see what your systems are actually doing
Software internals, explained visually.
Cloudmash takes the systems you build on every day, from Kubernetes and databases to distributed systems and LLMs, and follows one request through them, step by step, as animated diagrams.
- diagrams first
- one request at a time
- zero fluff
// topics
What we take apart
Five areas, one approach: open the box, follow the data, and draw every hop.
- 01
Kubernetes internals
What really runs when you type kubectl apply: the API server, etcd, watches, controllers, the scheduler and the kubelet.
- watch + informers
- reconcile loops
- scheduling
- 02
Distributed systems
Consensus, replication, partitions and clocks, drawn as messages moving between real nodes instead of proofs on a slide.
- Raft
- leader election
- exactly-once
- 03
Backend system design
How request paths, queues, caches and rate limiters fit together, and what breaks first when traffic spikes.
- queues
- caching
- backpressure
- 04
Databases and storage
The journey of a write: write-ahead logs, B-trees, LSM trees, MVCC and the indexes that make reads fast.
- WAL
- LSM vs B-tree
- MVCC
- 05
LLM and AI systems
What happens between hitting send and seeing tokens stream back: tokenizers, attention, KV caches, batching and serving.
- KV cache
- inference serving
- RAG pipelines
- ??
Your stack next?
Want a system taken apart? Leave a comment on any video with the component you want to see from the inside.
Suggest a topic
// latest videos
Fresh from the channel
All videos on YouTube-
Why AI Progress Is Moving Faster Than You Think
-
What Really Happens When You Hit Send on ChatGPT
-
DeepSeek vs Kimi vs Qwen: How FREE 2.8T AI Models Work
-
Why Does AI Fit on a $2 Chip but Need a Warehouse of GPUs?
-
Transformers Attention MHA, MQA, GQA, MLA
-
If You Understand This, You Understand Kubernetes
// latest posts
From the blog
All posts on cloudmash.blog-
How Kubernetes Works Internally: API Server, etcd, and the Watch Mechanism
What really happens when you run kubectl apply? Follow one Deployment through the API server, etcd, watches, informers, controllers, scheduler and kubelet.
// about
Why Cloudmash exists
Most tutorials stop at the interface: run this command, call this function, and it works. Cloudmash starts where they stop. Every video picks one real system and follows a single request, event or token through it, drawn as an animated flowchart where you can watch the data move between components.
The mission is simple: make the internals of modern infrastructure as clear as a whiteboard session with a senior engineer who has time to explain. If you have ever nodded along to "the controller reconciles the state" without picturing what that means, this is for you.
Videos live on YouTube. Written deep dives, with the diagrams and the code, live on cloudmash.blog.
- 01 / mechanism over magic
Every claim maps to something you could find in the source code or the docs.
- 02 / diagrams first
If it moves, it gets drawn. Watching a packet travel beats reading three paragraphs about it.
- 03 / depth, not hype
No hot takes and no "10 tools you must know". Just how things work, explained properly.