arXiv:2605.25297v2 Announce Type: replace-cross 
Abstract: Effective features are crucial for predictive model performance, but creating them often requires domain expertise, limiting scalability across applications. We define feature engineering as an agentic code generation problem: features are not static data transformations, but executable programs that can be generated, evaluated, and iteratively improved. We present Eureka, an LLM-driven framework with three stages. (1) An Expert Agent, fine-tuned via SFT on domain knowledge, produces structured feature design plans in JSON format. (2) An LLM Feature Factory translates each plan into executable Python code through chain-of-thought reasoning, turning feature hypotheses into runnable programs. (3) A Self-Evolving Alignment Engine uses Reinforcement Learning (GRPO) with dual-channel reward (metric-based utility + semantic alignment) to enhance code quality. By expressing features as programs, the learned generation patterns can transfer across domains. Evaluated on 7 public benchmarks in healthcare, finance, and social domains, Eureka consistently outperforms both traditional AutoFE and LLM-based baselines. We further demonstrate Eureka's effectiveness on cloud GPU resource demand prediction at Alibaba Cloud, where Eureka improves demand fulfillment rate by 16% and lowers computing resource migration rates by 33%.

قدمت Eureka إطارًا مبتكرًا لهندسة الميزات في توقع طلب موارد السحابة للذكاء الاصطناعي في المؤسسات، مستفيدة من عملية من ثلاث مراحل تشمل وكيل خبير، ومصنع ميزات LLM، ومحرك محاذاة ذاتي التطور. تعمل هذه الطريقة على تحويل تصميم الميزات إلى برامج قابلة للتنفيذ، مما يعزز قابلية التوسع وفعالية النماذج التنبؤية.

Eureka ha presentado un marco innovador para la ingeniería de características en la predicción de la demanda de recursos en la nube de IA empresarial, aprovechando un proceso de tres etapas que involucra un Agente Experto, una Fábrica de Características LLM y un Motor de Alineación Autoevolutivo. Este enfoque transforma el diseño de características en programas ejecutables, mejorando la escalabilidad y efectividad de los modelos predictivos.

Eureka a introduit un cadre innovant pour l'ingénierie des caractéristiques dans la prédiction de la demande de ressources cloud AI pour les entreprises, s'appuyant sur un processus en trois étapes impliquant un agent expert, une usine de fonctionnalités LLM et un moteur d'alignement auto-évolutif. Cette approche transforme la conception des caractéristiques en programmes exécutables, améliorant ainsi la scalabilité et l'efficacité des modèles prédictifs.

Eureka has introduced an innovative framework for feature engineering in enterprise AI cloud resource demand prediction, leveraging a three-stage process involving an Expert Agent, an LLM Feature Factory, and a Self-Evolving Alignment Engine. This approach transforms feature design into executable programs, enhancing the scalability and effectiveness of predictive models.