Open to 2027 new-grad MLE & DS roles

Jiashu (Jerry) Huang

I build machine-learning systems across computer vision, RAG, and applied decision science.

See my work
01 · selected work

Things I've built

Machine learning engineering — evaluation tooling, infrastructure, and reliability.

Evaluation tool · LLM systems

Schema Load

A reproducible CLI for testing whether OpenAI-compatible model services still return correct JSON under load. The runner varies concurrency, prompt and output length, temperature, seed, and structured-output mode.

$ schema-load report
schema_validity · field_correctness
latency · throughput · goodput
→ JSONL + CSV + Markdown, with confidence intervals
Role
Evaluation tooling and systems engineering
Focus
Structured-output reliability under load
  • LLM Evaluation
  • JSON Schema
  • Load Testing
  • Python
View on GitHub →
Research · Computer vision MLE + DS

Bee Wing Project

An interactive bee species identification system combining a ResNet raw-wing classifier with an interpretable geometry model built from wing venation and cell shape.

Bee Wing specimen workspace with a bee-wing image and staged geometry-model controls
specimen workspace → geometry model stages
90%ResNet accuracy
470+wing images
20+bee species
Role
ML research and engineering
Outcome
Two interpretable CV pipelines
  • Computer Vision
  • Image Segmentation
  • Interpretable ML
  • PyTorch
Try the live demo →
Mobility research · Forecasting MLE + DS

Dockside Evanston

A local research prototype that maps current Evanston bike and dock availability and estimates experimental 30-minute stockout risk. A Python service polls public GBFS feeds every minute and compares a state-transition model with a persistence baseline.

Dockside Evanston app: a list of Evanston bike stations with live bike counts beside a map, and a detail panel for Sheridan Rd & Noyes St showing 5 bikes, 6 open docks, under 1% chance of no bikes in 30 minutes, and a 24-hour station history chart
live app · 30-min outlook + station history
5,860labeled observations
6stations · 2026 validation pending
Target areas
  • MLOps
  • Scalable Training & Deployment
  • Production Readiness
  • Data Pipelines
View repository →
Decision science · Demand & revenue

Orlando Attraction Demand & Revenue Planning

An end-to-end planning system that turns daily attraction demand and revenue forecasts into pricing, campaign, and capacity actions. PostgreSQL and Python prepare the data and forecasts; a decision dashboard presents the backtest and next month's operating plan.

7.5%revenue WAPE
103.3ticket MAE
30-dayoperating plan
  1. Forecast daily ticket demand and revenue
  2. Backtest against held-out days
  3. Recommend pricing, campaign, and capacity actions
Pipeline
PostgreSQL, Python, Tableau
Outcome
30-day operating plan and decision dashboard
  • SQL
  • Demand Forecasting
  • Revenue Planning
  • Decision Science
View on GitHub →
Developer tool · Infrastructure as code

CDKCanvas

A browser-based visual editor for designing AWS infrastructure, validating resource connections, and generating TypeScript AWS CDK projects.

InfraCanvas visual AWS CDK builder showing connected Lambda, API Gateway, S3, and DynamoDB nodes
node canvas → validate → CDK .ts
Role
Product and frontend engineering
Outcome
TypeScript CDK generation with import and validation
  • AWS CDK
  • Infrastructure as Code
  • TypeScript
  • React Flow
View on GitHub →
Data product · Labor market

U.S. Workforce Intelligence Explorer

An interactive Tableau dashboard for exploring U.S. industries, occupations, skills, wages, and regional employment patterns using Bureau of Labor Statistics data.

U.S. Workforce Intelligence Explorer dashboard with wage, employment, and occupation-growth views
filters update every view
Role
Data engineering and visualization
Outcome
Four analysis-ready datasets from BLS source data
  • Tableau
  • Data Visualization
  • Python
  • ETL Pipelines
View on GitHub →
02 · experience

Where I've shipped

Hax.aiFull Stack Engineer

Full-stack engineer on a startup team

Built retrieval, evidence, and memory workflows for a collaborative AI product, deploying scalable application services and user-facing tools on AWS.

Cheadle Center for Biodiversity and Ecological RestorationResearch Assistant

Identifying bees from wing morphology

Built image-processing and feature-extraction pipelines for 470+ bee-wing images across 20+ species, and evaluated ResNet-50 and morphology-based classifiers for species identification.

Correkt StartupMachine Learning Engineer

Fact-checking with agents and retrieval

Developed a fact-checking system combining logical reasoning, a dynamic database, RAG, ReWOO, and LangGraph agents to improve real-time knowledge retrieval.

Xin Yi TechMachine Learning Engineer Intern

Productionizing vision models

Built and deployed video-origin classification and image-generation systems, fine-tuned large models, and exposed existing models through product-ready APIs.

03 · about

I'm interested in systems where model quality, infrastructure, and human inspection all matter. My work ranges from interpreting bee-wing morphology to building RAG workflows and demand-planning tools.

Tools
  • Python
  • SQL
  • AWS
  • REST APIs
  • PyTorch
  • LangGraph
Looking for

New Grad Machine Learning Engineer and Data Scientist roles.

Contact