Articles for people who
ship the stack
Rewrites on React, Node.js, TypeScript and AI. A translated body is shown when it exists. If it does not, you see the English text and a note that says so.
Tagged: Fine-Tuning
Training Enterprise AI Agents: Choosing Between SFT, DPO, GRPO and RAG
Training enterprise AI agents: when to use SFT, DPO, GRPO, LoRA or plain RAG, and how to ship the result as a pinned, evaluated and reversible release.
50 min readWhat a Model-Diffing Agent Finds: Style Shifts Overshadow a Narrow Fine-Tune
A model-diffing agent audited a finance LoRA fine-tune: 82% of runs caught the length and structure shift, while the intended reasoning change stayed hidden.
15 min readAGI Is a Learning Problem, Not a Storage Problem
An argument that continual learning, not bigger weights, is the missing piece of AI, with a sketch of a self-updating loop that validates each small change.
7 min readRAG, Fine-Tuning or Agents: Picking the Right Layer and the Order to Build It
RAG vs fine-tuning vs AI agents: what each does, cost, latency and failure modes, rules of thumb, and an HR assistant that shows the order to build them in.
12 min readFine-Tuning Got Cheap, but Agents Still Need a Gate for Weights
How to turn an agent's stored lessons into weights with on-policy self-distillation, four promotion criteria, and why repeating the loop three times breaks it.
9 min readPrompting, RAG or Fine-Tuning: Diagnose the Layer Before Choosing
A practical way to choose between prompting, RAG and fine-tuning in 2026, using three questions, a decision flow, cost shapes, freshness and common mistakes.
9 min readFrom Next-Token Prediction to Instruction-Following: Fine-Tuning a GPT Model
How a pretrained GPT learns to follow instructions: formatting records, label shifting, padding masks, the training loop, evaluation limits and the step toward DPO preferences.
9 min readSoft-Label Distillation for a 0.8B SQL Agent: Why a Weaker Qwen Teacher Won
LoRA fine-tuning on top-k soft labels from frozen traces took a 0.8B SQL agent to 55 of 220 tasks, and a weaker Qwen teacher beat the stronger GPT model.
20 min readGRPO Fine-Tuning Explained: Teaching a Local LLM to Reason
GRPO drops PPO's critic and learns from group comparisons. Walk through the five core quantities, the reward functions, and a short Unsloth and TRL training script.
11 min read
About these articles
Request a 24h estimate
Need the same stack in a production operator layer? Send the brief — estimate within 24 hours.