• Developed Large Language Model Agents using the AutoGen framework to evaluate LLM-generated answers;• Established a Logic, Accuracy, Relevance, and Completeness Module, and a Final Judge;• Constructed a grader and a challenger agent within each module, guiding the Final Judge to provide the ultimate grade; reduce the need for additional prompting to address vanilla LLM graders’ issues;• Achieved a 20% increase in grading accuracy, reducing the need for costly human intervention and evaluation.

• Utilized Average Treatment Effect, Granger time series causality test, and t-test to identify causality between FOMC chairman speeches and short term Foreign Exchange (FX) rate fluctuation;• Scraped 150+ FOMC chair speech scripts and 10-year FX historical rates between 5 currency pairs;• Fine-tuned LLAMA, BERT, AutoGen to predict monetary policy and short-term FX rate fluctuations.

• Used LLAMA to extract and process text from 6000+ county zoning documents;• Prompt-engineered ChatGPT to respond to inquiries about house price prediction and county zoning questions;• Performed PCA and T-SNE for visualization to identify different language styles of documents across states;

· Researched and gathered data about major institutional investments in public client companies and theircompetitors; created visualizations and conducted data analysis in Excel to present to senior management· Advised a PE firm on portfolio construction through comparing alpha-related data between portfolio companies and 20+ comparable companies, and recommending profitable portfolio allocation choices among investments· Utilized Excel VBA to build an automated financial tracker for a client to assess and analyze major investor holdings within their company, in addition to showcasing investor institution holdings in six major competitors

· Evaluated companies’ fundamental strengths by analyzing 10Ks, financial statements and prospectus· Conducted 10+ ad-hoc industry reports and company analysis and presented to firm’s investment committee· Analyzed the non-ferrous metal industry; forecasted underlying asset prices and conducted expert calls· Assisted senior directors prepare for roadshows; spearheaded comparable and precedent transaction analysis

· Drafted a 25+ page industry research on non-homogenous tokens (NFT) in Web3.0 to understand blockchain technology and cryptocurrency; interviewed 15+ users and experts to validate feasibility of decentralization· Sourced 20+ potential investments in Web3 weekly; utilized CrunchBase and PrivCo to compile financials