Factor Generation & AI Training Data for Quantitative Research
Document
Fine-tuningWorked on an AI-assisted quantitative factor generation and evaluation workflow focused on alpha signal discovery, factor expression generation, simulation review, and quality filtering. The project involved designing and testing candidate factor expressions, reviewing model-generated factor ideas, identifying invalid or low-quality signals, recording failed gates, and building reusable research memory to improve future factor generation. This experience is highly relevant to AI training data because it required evaluating model outputs, creating feedback signals, classifying failures, standardizing candidate quality, and turning research results into structured training/evaluation data. The workflow covered financial time-series data, market features, ranking operators, neutralization logic, turnover control, drawdown review, Sharpe/Fitness-style metrics, and automated candidate generation pipelines. I also worked with local templates, LLM-assisted generation, JSONL queues, simulation logs, rejection records, leaderboard reports, and iterative quality-control rules. This gives me strong experience in high-precision AI evaluation tasks where the output must be judged not only by language quality, but also by mathematical validity, domain logic, and downstream performance.
RLHF