LayerX

Control LLM output with temperature, top-p, penalties, and logprobs to ship reliable AI automation. A practical field guide with a per-task cheat sheet

How a 3-person LayerX team with no ML background won Fixathon, building Cadence: an explainable AI model that predicts factory production time from panel data.

Coding agents write 80% of the code. The other 20% is engineering, and it matters more than ever. Eleven lessons from building with agents at LayerX.

Taste differentiates your product. Trust determines whether anyone actually uses it for what matters.

Why the best opportunities in AI keep coming from rooms, not feeds. What two days at SIM Conference actually taught me.

The job of a senior marketer has shifted from managing people to designing systems.

AI is shifting from copilots to autopilots. Why selling outcomes beats selling tools, and how we're building LayerX for the Service as a Software era.

How we turned every LayerX blog post into an audiobook-style narration using open-source TTS models (Kokoro, Qwen3 TTS) running locally on Apple Silicon with MLX. No cloud, no per-character billing, no third-party data exposure.

The story of building Eva, a personal AI assistant that lives where my conversations already happen.

A hands-on guide to generating synthetic training data from FAQs, fine-tuning Qwen3-4B with LoRA on Apple Silicon, and serving the model locally with llama.cpp

After a week at EthCC Cannes 2026, a provocation: Web3's complex, unreadable interfaces aren't a UX failure, they're a feature.