How a convolutional network classifies an image
Convolution, activation, and pooling in the image classifier used by the demo.
12 min readWriting
How the fluid simulation works
Velocity, pressure, and the controls behind the fluid demo.
2 min readHow image kernels change a picture
Weighted sums, padding, and the filters used by the convolution demo.
2 min readInside a small language model
A walkthrough of the calculations in the small model used by the 3D visualization.
16 min readHow GPT-2 predicts the next token
Token embeddings, attention, and sampling in the GPT-2 visualization.
12 min readFive systems ideas worth touching
Small visual labs for consensus, indexes, embeddings, classifiers, and language-model sampling.
2 min readWhat the black-hole image is showing us
From curved light paths to sparse radio measurements: the geometry and reconstruction behind the EHT image.
5 min readA shrinking vortex and a growing velocity
Reading OpenAI’s Navier–Stokes construction through its core geometry, energy scaling and smooth forcing.
4 min readThe art of a boring deployment
Define rollout gates, preserve schema compatibility, and make rollback a routine operation.
4 min readYour RAG pipeline needs a test set
Measure retrieval recall per question and compare pipeline changes against a versioned evaluation set.
2 min read · NotebookEvery millisecond has a story
A Python notebook showing how a small slow cohort changes tail latency while the average stays low.
2 min read · NotebookAn alert is not an incident
Correlate alerts without losing source events, clock uncertainty, or the ability to reject a bad grouping.
2 min readSmall objects, surprising costs
Measure Python object overhead, choose the right storage representation, and account for conversion costs.
2 min readYour architecture rules deserve a linter
Use Go’s syntax tree to turn dependency direction, naming rules, and yesterday’s production bug into CI checks.
3 min readThe platform is a conversation
Design platform defaults around recurring developer tasks, explicit ownership, and measurable recovery.
2 min readA platform can be a library and a well-written manual
How a small SRE team used CDK constructs, operating guides, and guardrails instead of building an internal PaaS.
4 min readOrganize architecture docs by how fast they go stale
Separate the three-year decisions from next month’s tuning knobs, and the document becomes much easier to trust.
3 min readMCP, ACP, and A2A: which wire connects what?
Tool protocols and agent protocols solve different boundaries; the alphabet soup gets easier once the arrows are drawn.
3 min readA home lab built from the machines already in the cupboard
A practical plan for learning Rust and Kubernetes without pretending the first purchase needs to be a tiny data center.
3 min readA monolith with a BFF brain and a domain heart
Keep UI-shaped reads fast and domain-shaped writes honest, without deploying another service for sport.
3 min readVision Transformers, without the ceremonial fog
How an image becomes a sequence of patches, and why attention changed the computer-vision toolbox.
2 min readVLM or vision foundation model? Use the task to decide
Phi-3.5 Vision and Florence-2 look similar at the input, but they are built for rather different jobs.
3 min readGenerating Dify workflows with LangGraph
A small state machine that writes a workflow, checks its own work, and knows when to ask a human.
3 min readTurning apartment Wi-Fi into a site-to-site VPN
A Raspberry Pi, WireGuard, and one mildly hostile captive portal walk into an apartment network.
4 min read