Ziyan Xia 夏子言

Handyman — 3D repair guidance

Generative 3D hackathon · team project · handyman.ziyanxia.com

Sep 2026 · San Francisco, CA

• The problem. A repair guide tells you to tighten the hinge without telling you which part the hinge is. Handyman reconstructs a photographed object as a labeled 3D model you can explode, focus and isolate on screen, tying every step of a repair to a part you can actually see.
• Scope as a product decision. Small, low-risk repairs only, draft content labeled as draft, and an explicit stop-and-call-a-pro boundary for anything electrical, gas, structural or leaking — confidence without overconfidence.
• My part. Problem framing, product story and the pitch, built with a small team over the hackathon weekend. Full write-up on the project page.

• Generative 3D
• World Labs Marble
• Tripo
• Convex
• Next.js
• three.js

Data Analyst / Analytics Engineer

Revvo Technologies Inc. · AI road safety platform

2022 – 2026 · San Mateo, CA

• Metric design and product insights. Built end-to-end KPI frameworks for a large-scale IoT ecosystem (1B+ miles, 5K+ tire events), defining self-serve real-time metrics and customer-facing insights that enabled proactive road-safety alerts and contributed $75M+ in customer operating cost savings.
• Forecasting and machine learning. Deployed production regression, classification, clustering, and survival models for tire risk prediction, sensor positioning, and depot detection — up to 90% accuracy across 50+ enterprise fleets in 25+ U.S. states.
• Experimentation and causal inference. Designed and ran field A/B tests evaluating firmware updates, gateway repositioning, and signal-filtering algorithms, comparing telemetry distributions and downstream model outputs to prevent production regressions at scale.
• Anomaly detection and root cause analysis. Automated investigation of abnormal sensor behavior with Python/SQL diagnostics and cross-device comparisons, separating environmental noise from hardware faults and cutting false-positive tire replacements by 80%+ (200+/week to under 40/week).
• Production analytics workflows. Engineered analytics pipelines (Python, SQL/GCP, Java, Looker) behind Revvo's TireIQ platform, with Git, unit testing, and CI/CD for reliable deploys — investigation time down 75% (2 hrs to 30 min).
• Algorithm design and optimization. Designed sensor auto-registration algorithms using statistical methods and GPS telemetry, correcting noisy mappings and false positives and reducing installation calibration time 85% (5 days to 18 hrs).
• Cross-functional work. Partnered with Product, Engineering, and GTM teams to define success metrics, evaluate feature launches and account health, and deliver recommendations guiding product iteration, ML model calibration, LLM agent evaluation, and C-suite decisions.

• IoT
• B2B SaaS
• Experimentation
• Forecasting
• Looker
• GTM Partnership

Undergraduate Research Assistant

University of California, San Francisco · Roland Henry Lab
through the UC Berkeley URAP program

2020 – 2021 · San Francisco, CA

• Time-series forecasting and predictive analytics. Built end-to-end Python analytics and modeling pipelines on high-frequency wearable (Fitbit) clinical data, using ARIMA-regression hybrid models to forecast mobility trends and behavioral patterns across longitudinal patient cohorts and identify meaningful changes in mobility over time.
• Statistical analysis and data quality. Established subject-level baselines and deviation signals through within-subject analysis and hypothesis testing in Python, validated data quality, and identified anomalous observations, improving detection of meaningful motor-function changes over time.
• Clinical data processing and feature engineering. Cleaned and transformed longitudinal clinical data from disparate sources, implementing missing-value imputation, noise reduction, validation rules, and feature extraction to create reliable analytical datasets and stable inputs for downstream predictive modeling.
• Exploratory analytics and stakeholder-ready insights. Conducted exploratory analysis of clinical time-series data to uncover behavioral patterns, seasonal effects, and activity cycles, translating findings into interpretable analyses that supported hypothesis generation, model design, and understanding of longitudinal patient outcomes.

• Time Series
• Wearables
• Clinical Data
• ARIMA

Academic Standing

Central China Normal University

2017 – 2021

• Ranked in the top 10% of the Statistics cohort.
• First Prize, Brand C, 2019 National English Competition for College Students.

KAUST Fellowship

King Abdullah University of Science and Technology

2021

• Fully funded MS/PhD fellowship — tuition, stipend and housing — offered at a 5% acceptance rate.

• Analytics & experimentation. A/B testing, experiment design, causal inference, metric design, time series, LLM, NLP
• Programming. Python (NumPy, pandas, scikit-learn, statsmodels, TensorFlow, PyTorch), SQL, Java, R, JavaScript, Bash
• Data engineering. BigQuery/PostgreSQL, Spark, Airflow, ETL pipelines, batch processing, data modeling, API integration
• Visualization. Looker, Tableau, Plotly, Power BI, R Shiny, Excel/Google Sheets
• Production practices. Git, Docker, CI/CD, unit testing, Agile/Scrum (Jira)

Bike Share Demand & Rebalancing

Predictive modeling · Washington, D.C.

• Predictive model for bike availability, used to inform reshuffling strategy across the city from multi-year usage data.

Fitbit peak detection & DTW clustering

Time series · R

• Peak detection on Fitbit activity data, with dynamic time warping used to cluster daily activity patterns.

Dockerized Python functions for PostgreSQL

Data engineering · Python

• Walkthrough of deploying Python functions for PostgreSQL inside Docker containers.

XGBoost vs. iterative Random Forest on imbalanced high-dimensional genetic data

Write-up (PDF)

• Comparison on imbalanced, high-dimensional genetic datasets; iterative Random Forest outperformed XGBoost.

Social and Behavioral Research — Basic/Refresher

CITI Program

Issued Feb 2022 · Expired Feb 2025 · ID 47150531

English Teacher Volunteer

Arsasom Organization (Thailand NGO)

Jan 2019

• Taught English and created instructional media at Phasukmaneejakmittraphab 116 School, a public school in Thailand.