Tools
References I needed, so I built them.
Interactive guides to decisions that come up constantly in AI infrastructure work. Checked against primary sources rather than assembled from blog posts.
Fine-tuning cost calculator
Estimate a fine-tuning project for text, images, videos, audio or classifiers. Choose your examples and delivery package to see a simple price range.
Build an estimate →Fine-tuning method explorer
Twenty-nine methods — LoRA and its variants, memory tricks, alignment and multi-GPU strategies — with a wizard that narrows by hardware and constraint.
Open the explorer →Inference engine explorer
Thirty-one serving and local engines with verified repo links, and a configurator that picks one for your model size, GPUs and use case.
Open the explorer →Will it fit? — VRAM calculator
Model size, quantization and context to a VRAM figure — weights, KV cache and optimizer broken out — and how many 3090/4090/5090/A100/H100 GPUs that needs.
Open the calculator →GPU comparison
NVIDIA, AMD and Intel — RTX 3090 up to Blackwell and CDNA4 — VRAM, bandwidth, FP16 TFLOPS, low-precision and power. Filter by vendor, sort any column.
Open the table →Quantization format explorer
GGUF, AWQ, GPTQ, EXL, FP8, FP4, NF4 — size versus quality and which engines read each, so you can pick the right bits-per-weight.
Open the explorer →The VC Framework
A disciplined process for building software with agentic coding CLIs — Claude Code, Codex, Gemini CLI — plus what changed and what's timeless. Refreshed July 2026.
Read it →Something missing?
Tell me what you'd want to look up.
These started as notes I kept re-deriving. If there's a decision you keep making from scratch, it might be worth building.