My AI Resume

▸ LLM-Powered Industry Rotation Advisory Agent

• •Collaborated cross-functionally with product, engineering, and data teams to architect module structure, data pipelines, and agent workflow for an LLM-powered industry rotation advisory agent; authored PRD covering six analytical dimensions (macro, valuation, sentiment, technical, capital flow, and crowding)
• •Partnered with data infrastructure team to define and integrate API requirements for 20+ financial indicators (e.g., sector P/E percentiles, northbound fund flows, turnover rate time series), specifying data types, historical depth, update frequency, and granularity to support real-time agent inference
• •Iteratively refined prompt engineering strategy and report architecture; redesigned output to a conclusion-first format and developed structured extraction prompts to parse unstructured sell-side research reports into actionable investment signals across six dimensions, improving output consistency for sector rotation recommendations

▸ AI Intelligent Customer Service Chatbot

• •Assisted in developing an RAG-powered chatbot for a leading financial services firm by automating document processing, utilizing MongoDB for metadata, Azure Cognitive Search for vector databases, and Azure CosmosDB for the knowledge graph
• •Authored and integrated a reranking function by embedding the ms-marco-TinyBERT-L-2-v2 model, improving response accuracy by 20%
• •Assisted in testing and deploying the model on the chatbot's API server

▸ AI Intelligent Invoice Recognition System

• •Engineered a computer vision AI system for intelligent invoice recognition, capable of identifying both handwritten and printed text
• •Led the dataset collection and data labeling process for fine-tuning a YOLOv12m-based object detection model
• •Assisted in parameter tuning for the model and implemented distinct OCR processes for handwritten and printed fonts post-detection

▸ AI Investment Portfolio Management Assistant

• •Developed an AI assistant for investment portfolio management, featuring a conversational agent with tool-calling capabilities and a predictive module
• •Utilized an AI agent to dynamically invoke tools from an MCP server, enabling real-time data access and search functionality for a financial chatbot
• •Trained an LSTM model on NVIDIA L20S chips using historical minute-by-minute stock price data, achieving 15-minute price predictions with about 60-70% directional accuracy