Résumé

Education

Research projects adamzhao.dev/projects

• Built a simulator of complete company books (5 business types) with 15 planted reconciliation, fraud and prompt-injection problems, clean-twin controls, and a scorer built on the automation rate at 1% error; a rules baseline decides 96.5% of a month's items unattended with zero false alarms.

• Analyzed 1.69M SBA 7(a) loans: bunching at the guarantee cutoff (replicating Bachas, Kim and Yannelis: elasticity 3.9 vs 4.6), regression discontinuity under manipulation, a competing-risks default model (out-of-time AUC 0.68 to 0.80) and a Monte Carlo stress test; found and removed a program-ceiling confound and a post-default data leak.

• Built a point-in-time SEC XBRL warehouse (315 firms, 4.9M facts) with accounting-identity checks, and a likelihood-based memory test for language models with known training cutoffs (6,868 firm-quarters, 8 models).

Experience

• Reconstructed RMB 265M (about USD 37M) of fund flows and shareholder transactions from two years of raw bank records: cleaned, de-duplicated and matched transfers across accounts, then mapped counterparties into a shareholder ledger that became key financial evidence in a corporate dispute.
• Built a regression model of office-property values and ran depreciation and downside scenarios on the firm's real-estate exposure, quantifying RMB 40M (about USD 5.6M) of potential losses for the strategy review.

• Built a CRM-enabled outreach workflow (segmented lead lists, email sequencing, LinkedIn touches, response tracking) for a Japanese enterprise-software client entering the US; online lead generation doubled over the engagement.
• Built risk-factor profiles of historical Japanese software entrants and current US prospects; roughly 10% overlap in key structural factors supported the market-entry recommendation.

• Co-developed a weighted-average investment risk index separating policy, market and execution factors, and applied it to overseas energy projects for four clients, including a USD 400M Brazil investment completed in 2025.
• Synthesized intellectual-property and workforce data into an internal-structure optimization for an energy client and presented the recommendations to client leadership.

Earlier projects

• Managed a simulated USD 100K portfolio across 9 positions screened from 20 to 30 ETFs; rebalanced through the semester and finished 1st among participating teams.

• Merged 61 years of football, coaching and NFL data; benchmarked 7 models with 5-fold cross-validation; gradient boosting reached 64% out-of-sample accuracy.

Skills