42 articles
Series Years after building the original ML pipeline, LLMs changed everything — using the ChatGPT API to simplify elaborate recipes into weeknight-friendly meals, and reflecting on a decade-long journey from cosine similarity to LSTM to GPT.
Series A deep dive into LoRA, quantization, QLoRA, and SFTTrainer -- the key techniques that enable fine-tuning large language models on consumer hardware, validated through a Mount Fuji proof-of-concept experiment.
Series How an experiment where ChatGPT outperformed professional fund managers inspired me to build Senrigan, an AI stock price prediction service powered by a fine-tuned LLM.
Series My first attempt at fine-tuning ELYZA Llama-3-JP-8B on a local RTX 3060 GPU, and the hard lessons learned about VRAM limitations when training large language models.
Series Fine-tuning ELYZA 8B and LLM-jp 7.2B for stock price prediction on Google Colab, the accuracy challenges encountered, and why I ultimately pivoted to the OpenAI API.
Series How pivoting to OpenAI API fine-tuning on gpt-4o-mini achieved stable JSON output and sufficient accuracy in just 8 minutes, after months of struggle with open-source models.
Series A detailed look at how I designed training data for stock price prediction, integrating company information, news, stock prices, financials, and macroeconomic indicators into a structured JSON format for LLM fine-tuning.
Series How choosing DeepSeek as a translation provider based on cost alone led to Chinese text leakage, latency issues, and data exposure -- and why I unified everything under ChatGPT.
Series The challenges of migrating from MySQL to Firestore for a production web service, including index design, upsert strategies, document ID design, and cost optimization for the Senrigan stock prediction platform.