← Blog
docteur-turboss
I Audited My AI Coding Agents Instead of Trusting Them Blindly
How two audits written for the 4th technical report of my trading platform revealed two distinct failure modes: an agent that forgets because the repo no longer fits in its context window, and an agent that writes ten times more code than needed because nothing pushes it toward conciseness.
How Machines Talk to Each Other: An Overview from TCP to mTLS
Why TCP, UDP, TLS, mTLS, HTTP, and WebSocket are not competing alternatives but stacked layers; a hierarchical overview of machine-to-machine communication, from raw transport to mutual authentication.
I Evolved a Neural Network Through Natural Selection Instead of Gradient Descent
How I replaced classic gradient descent training with a NSGA-II genetic algorithm to evolve DQN trading agents: four versions, from overfitting to Lamarckian weight evolution.