Technical writing
Tutorials and articles on AI engineering
Tutorials explain one idea step by step. Articles document experiments, benchmarks, and implementation notes.
Tutorials
Start with the idea, then make it concrete with code.
KV Cache in LLMs From Zero
How KV cache changes LLM decoding, why it saves compute, and where memory becomes the bottleneck.
Speculative Decoding From Zero
How draft tokens are proposed, verified, accepted or rejected, and why the speedup depends on acceptance rate.
OpenAI Triton From Zero
How Triton kernels think in tensor blocks, with examples for masking, fusion, softmax, matmul, and benchmarking.
Articles
Experiments, benchmarks, and implementation notes.
Building Netron OCR: Four Models, One Screenshot
How I turned photographed Netron graphs into editable JSON with four compact vision models, a synthetic camera simulator, and a hand-annotated real-world evaluation set.
How to Squeeze More Juice Out of Coding Agents Without Burning Money
Context engineering practices for keeping coding agents useful without dragging every file, log, and failed attempt through the next model call.
Read articleWhat Actually Speeds Up Transformer Inference?
Profiling and optimizing a small autoregressive transformer with JAX, KV caching, batching, graph compilation, and low-bit inference.
Training a 10M-Parameter Transformer to Learn 3-Digit Arithmetic
A code-first experiment that builds a character-level decoder-only transformer with JAX, Flax, and Optax, then trains it on generated addition, subtraction, multiplication, integer division, and modulo problems.