AI & Business Analytics
Alibaba International
AI Business Analytics Intern
May – Aug 2026China
Joined Alibaba International's BI team supporting AliExpress logistics and supply chain. I combined causal analysis, semantic data infrastructure, and AI agents to quantify the business value of delivery improvements, standardise metric definitions, and accelerate the diagnosis of global logistics exceptions.
$1.27M
Monthly GMV opportunity quantified
1.4M
Users in delivery analysis
6
Delivery stages monitored
30 sec
Root-cause drill-down
Fulfilment & GMV Analysis
- Designed a three-window cohort study for US cross-border users and applied Generalized Propensity Score (GPS) with Difference-in-Differences (DID) to estimate the causal relationship between delivery time and subsequent GMV.
- Identified an approximately 10-day tolerance threshold: delivery delays beyond it were associated with accelerating GMV loss, with the steepest impact among high- and ultra-high-value users.
- Estimated that shortening delivery for users above 12 days to 10 days could unlock approximately $40.8K in daily GMV and $1.265M monthly, with 83% of the gain contributed by high-value segments.
AI-native Metrics Library
- Built an AI-readable logistics and supply-chain metrics layer using Markdown and YAML, combining a searchable index, standardised metric definitions, calculation logic, data lineage, and a business glossary.
- Created reusable Skills that extract metric definitions from reports and trace SQL from result tables through intermediate layers to ADS source tables, producing field-level lineage reports.
- Designed an agent workflow for metric lookup, document-definition audits, and change requests, connected through Devix and DingTalk for cloud-hosted, version-controlled access.
Logistics Sentinel Agent
- Developed a daily monitoring workflow that scans six fulfilment stages across countries and compares performance with SLA targets, seven-day trends, and thirty-day baselines before pushing exception summaries to DingTalk.
- Designed a three-level diagnostic path from country and fulfilment stage to the ten weakest routes and their lowest-performing carriers, reducing root-cause drill-down to about 30 seconds.
- Extended diagnosis with bilingual web attribution, matching geography and event timing to assess external causes such as port closures, strikes, and aviation controls.
AI & GPU Capability Building
- Authored an AI Workbench operations guide covering repository setup, ODPS connectivity, dependency installation, debugging, and grey-release deployment to help the team operationalise reusable analyses.
- Deployed and tested NVIDIA cuOpt with Qwen2.5-7B for vehicle-routing optimisation, translating business constraints into solver-ready models and converting GPU results back into decision-oriented explanations.
Tools & Stack
PythonSQLODPSGPSDIDYAMLDevixDingTalkQwen2.5-7BNVIDIA cuOpt