Duke University × Top-Tier Foundation Model Company
Co-developed quality control for more than 300 contributors across travel, food delivery, consumer goods, and e-commerce advertising. Built reliability-based review rules, expert adjudication, and random audits; audit records showed a higher share of outputs meeting every user requirement under the combined process. Extended this work into a mathematical design connecting residual-corrected quality estimation, audit allocation, and reviewer capacity. In a 10,000-submission synthetic experiment with ideal residual calibration, the allocation reduced estimation variance by 17.79% versus uniform sampling at equal expected audit cost.
Duke University × Top-Tier Foundation Model Company
Led financial and business task design for a frontier foundation model company, defining the task environment, reference cases, synthetic expansion rules, and expert preference protocols for post-training and evaluation. Guided experts with doctoral-level subject expertise and created Batch Zero to test independent execution before scaling; the company adopted it for other new tasks. The accompanying analysis derives constraint-preserving generation methods and quantifies how input changes, future information, and related task variants affect labels and evaluation.
Developed original physics, mathematics, and risk-engineering tasks, complete solutions, and failure feedback for a frontier AI research laboratory. Built acceptance methods around explicit proof requirements and valid alternative solutions. The accompanying analysis develops proof obligations, error certificates, and controlled task transformations. In an exact-arithmetic stress test of 3,600 constructed submissions across 600 quadratic problem families, semantic certificate checking accepted all 1,200 valid submissions and rejected all 2,400 invalid ones, while retaining alternative proofs that strict reference-path matching rejected.
Duke University × Leading Global Alternative Asset Manager
Developed an investment-research workflow that carried corrections to company evidence through scores, predictions, rankings, and explanations in a university-industry collaboration. Combined structured scoring, AI-assisted verification, tree models, SHAP, and stability analysis to make conclusions traceable and repeatable. The accompanying mathematical analysis represents evidence uncertainty and model choice jointly, derives conditions for certifying company comparisons, and prioritizes verification by the decision regret a query can remove. Constructed examples show how shared errors cancel and targeted verification resolves uncertain rankings.
Owned business development for Jiritsu's asset-verification and privacy-preserving computation services as the sole dedicated contributor for new issuers. Reported directly to a co-founder, combining data engineering, preference ranking, and counterparty discussions to develop partnership recommendations for the board.
Led the China workstream of a global market-entry engagement covering China, Japan, and the Middle East for a premium European appliance brand. Combined market data, consumer evidence, and policy research to recommend local products, pricing, channels, and marketing while maintaining the client's global positioning.
Applied statistical methods to sustainability assessment and green-credit screening within China Construction Bank's internal framework. Combined company data, historical assessments, and explicit eligibility rules to support candidate review, then documented the method for product managers and risk officers evaluating its use on a platform.
Duke Law School: Space Law and Autonomous Systems Research
Developed a legal framework for authorizing autonomous space operations when delayed human intervention creates risks of its own. Formulated a two-threshold test separating the authority to act from the evidence needed to delegate execution. Applied it to delayed spacecraft operations, orbital avoidance, debris capture, and SETI, then proposed task-specific authorization records. The resulting framework identifies which decisions can be authorized in advance, which require renewed review, and what evidence must support each choice.
Duke Law School: Space Law and Strategic Infrastructure Research
Developed a comparative legal framework for assessing small states' control over foreign-dependent space infrastructure. Analyzed New Zealand's launch regulation, Norway's Arctic satellite partnership, and Luxembourg's LUXEOSys procurement to distinguish permission, ownership, operational rights, and implementation capacity. Proposed an accountable-acceptance standard for strategic dependencies, linking interruption powers and response rights to funded capability. The analysis identifies specific institutional remedies for licensing, service continuity, procurement oversight, and collective action.
Duke Law School: Climate Law and Geoengineering Governance Research
Developed two proposed legal duties for solar geoengineering: comparative justification and continuity. Connected risk-risk analysis to authorization by requiring proponents to defend policy alternatives and answer challenges from affected constituencies. Traced termination risk to a separation between political control, operator succession, and the physical cooling trajectory. The resulting framework specifies conditions for deployment and withdrawal, supports governed research, and requires continuing assessment of both intervention and nondeployment.
Duke Law School: Environmental Law and Ecological Restoration Research
Developed a comparative legal framework for deciding when ecological restoration can justify new habitat loss. Distinguished protection, repair, and compensatory mitigation across Florida and the Greater Bay Area, using West Lake, Deep Bay, and San Tin to examine evidence of functional replacement. Proposed site-specific baselines, staged authorization and ecological acceptance, funded correction, and continuing responsibility. The framework connects restoration evidence to development permissions, funding decisions, and the point at which obligations can be discharged.
Served as the project team's ESG lead on acquisition advisory for a Southeast Asian conglomerate seeking Chinese office properties. Worked with JLL's internal ESG team to connect asset screening, property valuation, and site-level diligence with the client's investment requirements and potential refurbishment or repositioning plans.
Led shipping, logistics, and industrial asset-tracking research for Hubble's Bluetooth satellite commercialization through a Duke Practicum. Assessed where the technology's cost, accuracy, power consumption, and periodic reporting capabilities matched customer needs, translating technical and market evidence into recommendations for early applications.
Owned public-data research and governance screening for a Greater China dental-investment theme. Worked with the VP and investment team to prioritize subsectors and candidates, then translated governance concerns into investment recommendations, management questions, and on-site diligence priorities.
Supported the Agriculture Research Group's sector analysis and research-report work, contributing evidence for internal investment-strategy discussions.
Independently coached students preparing for advanced physics Olympiads through online and in-person instruction, combining subject teaching with competition problem-solving practice.
The High School Attached to Hunan Normal University
Trained in physics and mathematics Olympiads and served as a peer mentor in the school's nationally leading physics competition program, combining advanced study with collaborative problem solving.