Hi, I'm

Built a Douyin seeding analysis framework for Toread's new outdoor-casual product line, linking competitor landscape, campaign effectiveness, creator tiering, and content pattern analysis across Camel, Under Armour, Kailas, and Beneunder-style outdoor competitors. Using a cost-effectiveness-efficiency KPI system, peer benchmarking, and a creator evaluation model, I identified the highest-leverage combination as mid-tier creators plus visually atmospheric, low-ad-feel content, then translated the findings into reusable creator selection rules, content templates, and stage-based campaign evaluation logic for lower-tier market expansion.

Built a holiday demand forecasting and operations monitoring solution for Spring Festival demand fluctuations. Integrated multi-source data and applied machine learning with historical trends, seasonality, and holiday effects to forecast assurance-period workloads. Established dynamic-threshold monitoring and anomaly alerts for core KPIs, then optimized upgrade rules, priorities, and cross-channel collaboration between outsourced and in-house teams. During the Spring Festival assurance period, demand intake increased by 12% YoY while achieving 0 energy dispatch service accountable incidents, 0 complaints, and full compliance on key SLAs; quantified forecast-bias causes and created reusable data cases for future holiday model optimization.

Built an operations monitoring dashboard for the energy dispatch command team to address fragmented service scenarios, inconsistent manual records, and limited visibility into process efficiency. I mapped core workflows including battery monitoring, image recognition, BaaS approvals, the 9903 hotline, and home/public charging services, then integrated ticketing, hotline, and workflow data into Datasight. The dashboard defines metrics across workload, response efficiency, handling quality, and repeat contacts, with trend monitoring, drill-down tables, and reason-distribution views to support anomaly diagnosis, staffing evaluation, strategy review, and agent performance improvement under a unified metric standard.

A mixed-methods study of adults aged 60-69 in Hangzhou, combining surveys, employer interviews, policy text mining, and case analysis to evaluate re-employment intentions and the effectiveness of existing policy support. Drawing on 61 policy documents and 982 survey and fieldwork records, I used factor analysis and logistic regression to identify education support, employment services, and market mechanisms as key drivers, then translated the findings into targeted policy recommendations for highly educated older adults and older migrants living with their children.

Developed a Napari-based biomedical image annotation plugin for Charles Perkins Centre that combines Micro-SAM with interactive correction for multi-channel microscopy workflows. Implemented folder-level batch loading, per-channel embedding cache, auto segmentation, point/box/brush/eraser refinement, committed mask management, and TIFF/ZARR export to support reproducible downstream analysis. Validated the tool through repeated functional, usability, and performance testing, delivering stable segmentation results across nucleus, lipid-droplet, and other cell-structure annotation tasks.

Built a real-time fuel price monitoring platform on top of the NSW FuelCheck API, MQTT, and a Streamlit/Folium dashboard, covering the full workflow from data retrieval and cleaning to stream publishing and map-based visualization. The back end fetches live price data every 60 seconds, performs type conversion, deduplication, and station-level merging, then publishes cleaned records to MQTT at 0.1-second intervals; the front end subscribes to the stream in real time and supports filtering by fuel type, brand, and price range, along with live map markers, summary statistics, clustering, CSV export, and stable high-frequency updates through queue-based consumption, debounced rerenders, and preserved map state.

Built an advertising analytics workflow for health and nutrition products by integrating 25M+ exposure, ad, audience, and transaction records with SQL/Tableau Prep, including data cleaning, field standardization, and wide-table modeling. Designed dashboards for GMV, CTR, CVR, media cost, and segment conversion with campaign / product / channel / audience drill-downs; applied k-means clustering and chi-square testing to identify price-sensitive segments, inefficient channels, and high-potential content patterns, supporting budget allocation, media pacing, and campaign postmortems.