SleepyQuant
A public notebook from a solo finance + tech enthusiast — multi-agent quant on Apple Silicon, all local, all transparent.
Join the ride — see me fall or thrive, whichever comes first.
📬 The public build log
Solo builder running a 12-agent AI quant on one M1 Max. Every Friday: the architecture that broke, the trades that worked, the freezes. Real numbers, both directions. No hype, no signals, no paid tiers.
One email a week. Unsubscribe anytime with one click.
Multi-Agent System
A handful of specialized agents — trading, research, risk — sharing a single MoE model through a priority queue. Hand-rolled, occasionally fragile.
Transparent PnL
Every round-trip logged. Every loss written up. Paper first, real capital only when the math earns it.
Local-First
One Apple Silicon machine. 64 GB of unified memory. Zero cloud inference, zero API bills, zero dependencies I don't own.
Build in Public
Every decision, every failure, every architecture note shared openly. Nothing hidden, nothing polished.
From the notebook
Why Apple Silicon Quietly Won the Local-AI Race (April 2026)
While the public AI narrative is dominated by capex wars and cloud GPU shortages, a quieter shift has happened on the desktop. A single Apple Silicon laptop with 64GB of unified…
The Inverted Control: What 24 Hours of Running Our Own Bot Backwards Revealed
After roughly 500 paper round-trips showed a persistent sub-35% win rate with average losses larger than average wins, we stopped scaling the live side and ran a cheap experiment:…
I Blamed the Model for Months. The Bug Was My Sampler.
Running local LLMs on M1 Max hardware is one of those setups that looks great on paper — unified memory, no PCIe bottleneck, offline and private. For about a year I ran…
How I Budget 64 GB Unified Memory on M1 Max for a 35B Model + Long-Running Agent Loops
The first lie I had to unlearn buying a 64 GB Mac for local LLM work was that I had 64 GB to use for the model.
MLX vs llama.cpp on M1 Max with 35B Q8 — The Honest Benchmark
I tested both. Same machine (M1 Max 64 GB), same model (Qwen 3.6 35B-A3B Q8), same prompts, same generation lengths. llama.cpp came out about 30% faster on raw decode throughput.…
What to expect
- ✓ One email per week — short, data-dense, sent when there's something worth reading
- ✓ Real numbers — PnL, win rate, drawdown. No polished screenshots, no cherry-picked weeks
- ✓ Failure post-mortems — the trades that blew up, and what I missed
- ✓ Architecture notes — local LLMs on one Mac, multi-agent orchestration, the stuff that broke before it worked
- ✓ Zero hype — no "X coins to 100x", no signals to buy, no paid groups. Just the math
Looking for older posts? Read the full archive →