Making CEL faster: from AST interpreter to compiled closures
May 27, 2026 · 10 min · Maxime Guerreiro
RTX 5090 power scaling: 450W vs 575W training
May 21, 2026 · 7 min · Maxime Guerreiro
How I use agents to write this blog
May 21, 2026 · 5 min · Maxime Guerreiro
Soft distillation vs. gradient boosting on fraud
May 20, 2026 · 15 min · Maxime Guerreiro
Replicating Talking Trees: LLMs for fraud detection
May 20, 2026 · 6 min · Maxime Guerreiro
Allocator shootout for async Rust on ARM64
May 19, 2026 · 14 min · Maxime Guerreiro
Fine-tuning TabICL: when 30 epochs of GPU time buys you 0.3 pp
May 18, 2026 · 8 min · Maxime Guerreiro
Agent architecture: where the work runs
May 18, 2026 · 5 min · Maxime Guerreiro
When stacking works: it depends on which features your models look at
May 17, 2026 · 15 min · Maxime Guerreiro
TabPFN3 vs TabICL: a matched-size fraud-benchmark sweep
May 16, 2026 · 15 min · Maxime Guerreiro
Inlining Tokio MPSC recv: removing the async tax
May 16, 2026 · 3 min · Maxime Guerreiro
Tokio MPSC Sweep: message size vs latency
May 15, 2026 · 2 min · Maxime Guerreiro
Amortizing tokio’s global queue acquisitions
May 14, 2026 · 4 min · Maxime Guerreiro