Writing

Deep Dive

Claude Code Treats Its System Prompt Like Infrastructure

I read Claude Code's source looking for prompt engineering. I found cost engineering. Cache boundaries, sticky latches, circuit breakers, and the 77% number.

agentsclaude-codeprompt-engineeringcost-engineering
Experiment

Watch a Language Model Think

I used sparse autoencoders to look inside GPT-2. I found interpretable concepts, feature loops, and a temperature feature that fires correctly while the model gets the answer wrong.

interpretabilitySAEinteractive
EXP A Single Neuron for 'And' in Six Languages
Deep Dive

Why I'm Building Trust Bench

Evaluation tools score LLM outputs. They don't tell you why models fail. Trust Bench connects profiling, diagnosis, and repair into a single open-source tool.

trust-benchllm-safetyevaluationinterpretability
TOOL Three Behavioral Evals on Claude's Medical Safety
TOOL The Medical AI Evaluation Problem
EXP When an LLM Says 90%, Should You Believe It?
DEEP DIVE Six Experiments That Built My Intuition for Trust Bench
Tool

I Built a Terminal Dashboard to See Where My Tokens Go

crux is a TUI that reads Claude Code session logs and shows you real-time context growth, cache efficiency, cost breakdowns, and session health grades.

rusttuiclaude-codetoolingopen-source
brew install amaljithkuttamath/tap/crux cargo install crux-cli
TOOL I Built a Tool to Fix Claude Code Skills
TOOL Comparing LLM Tokenizers Side by Side
TOOL What's Actually Inside a GGUF File?
Experiment

Hybrid Attention at 8M Params (and What Broke at 150M)

I trained Qwen3.5's hybrid DeltaNet+attention architecture from scratch on a MacBook. Pure attention won at 8M. Scaling to 150M hit a math bug that looked like a hyperparameter problem.

mlxattentiondeltanettrainingapple-silicon