Blog
Deep dives, engineering stories, real-world use cases, industry shifts, and practical lessons from the world of AI.
Not tutorials. Not daily news. Stories worth understanding.
AI Engineering
RAG is often explained as giving an LLM extra information. That misses the harder engineering problem: finding the right evidence from everything you could possibly give it.
◷ 10 min read
Read the full article →
How AI Works
A slow, number-by-number explanation of attention—from context and Query, Key and Value vectors to dot products, masking, softmax, multi-head attention and KV cache.
◷ 21 min read
How AI Works
Follow one page from training data into tokens, gradients and model weights—and then watch those learned parameters answer a simple question.
◷ 18 min read
How AI Works
A slow, beginner-first explanation of underfitting, overfitting and generalization, with training curves, examples, diagnosis, fixes and modern AI connections.
◷ 20 min read
How AI Works
A slow, number-by-number journey through backpropagation—from a loss of 0.782 to gradients, updated weights and a better second prediction.
◷ 14 min read
How AI Works
A slow, number-by-number journey through one forward pass—from ‘The sky is ...’ to a prediction, a probability and a loss.
◷ 15 min read
How AI Works
A beginner-first explanation of gradient descent, learning rates, loss landscapes, batches, momentum, Adam and AdamW—continuing our neural-network example.
◷ 14 min read
How AI Works
A complete beginner-first walkthrough of one next-token training step—from text and vectors to logits, softmax, cross-entropy, gradients, backpropagation and a real weight update.
◷ 24 min read
How AI Works
A beginner-first debug trace of one request through an AI assistant—from the browser and safety layer to tokens, vectors, attention, logits, sampling and streamed output.
◷ 29 min read
How AI Works
GPUs did not win AI because they are simply faster processors. They won because neural networks contain enormous amounts of mathematical work that can happen in parallel.
◷ 7 min read
Agents & LLMs
AI agents can search, use tools, change plans and take actions. Underneath the magic is a surprisingly understandable software loop.
◷ 7 min read
AI Engineering
Parameter count once dominated the AI conversation. Production AI is forcing a more useful question: how much capability do you need for this task, at this latency, on this hardware, for this cost?
◷ 7 min read
AI Industry
A prompt feels weightless. The infrastructure behind it is not. Follow one AI request backward from the chat box to GPUs, networks, cooling systems and the electrical grid.
◷ 7 min read
Real-World AI
The important shift is not 'AI writes code.' Developers are moving from producing every line manually toward specifying, reviewing, testing and supervising increasingly capable coding agents.
◷ 7 min read