Jiamu
Chen陈佳木

Research Assistant · High-Frequency Prediction of SPX Returns and Spreads under FOMC Meetings研究助理 · FOMC 会议期间标普 500 指数收益与价差高频预测

Built an LLM pipeline to extract sentiment and ambiguity signals from 15 years of FOMC meeting videos, combining them with order-book and transaction data to construct second-level market features.构建大语言模型流程，从 15 年的 FOMC 会议视频中提取情绪与模糊性信号，并结合订单簿及成交数据构造秒级市场特征。

Designed a Sparse-Group LASSO framework with Legendre polynomial transformations to jointly forecast SPX returns and bid–ask spreads. Also built an XGBoost workflow to nowcast firm-level P/E ratios and contributed to a financial agent integrating structured market data with unstructured text.设计结合勒让德多项式变换的 Sparse-Group LASSO 框架，联合预测标普 500 指数收益与买卖价差；另构建基于 XGBoost 的公司市盈率预测流程，并参与融合结构化市场数据与非结构化文本的金融智能体开发。

Quant Research Intern量化研究实习生

Mined 60+ predictive signals across daily, minute, and tick data. Built PyTorch models to uncover 20+ novel factors, screened 4,000 factors, and combined 200 selected factors into portfolio signals.基于日频、分钟与逐笔数据挖掘 60 多个预测信号；使用 PyTorch 模型发现 20 多个新因子，从 4,000 个因子中筛选并组合 200 个因子构建组合信号。

Used NVIDIA RAPIDS, multiprocessing, and vectorized computation to reduce feature-extraction latency by 50%.运用 NVIDIA RAPIDS、并行多进程与向量化计算，将特征提取延迟降低 50%。

Metrics as reported in the resume.指标依据简历所列研究结果。

Graduate Teaching Assistant研究生助教

Supported Machine Learning for Finance and Numerical Methods for Quantitative Finance, including stochastic modeling, numerical algorithms, and project coding.协助《金融机器学习》与《量化金融数值方法》课程教学，内容包括随机建模、数值算法与项目编程。

Project Contributor · KVCache and ToolCache Codesign项目贡献者 · KVCache 与 ToolCache 协同设计

Helped define a tool-augmented LLM serving problem and proposed a cache-aware framework balancing KV reuse, tool-result reuse, eviction policy, memory, and latency.参与定义工具增强型大语言模型服务问题，并提出缓存感知框架，统筹 KV 复用、工具结果复用、淘汰策略、内存与延迟权衡。