Home

Complete archive

Systems, models, products, and the work between them.

Software engineering leads the story. Applied machine learning, research, robotics, and community work show how the same systems thinking travels across domains.

Experience

8 roles

Logo preview

Logo paths are optional in the experience data.
01

Jun 2026 - Present

San Diego, CA

AI Quality Engineer Intern

Dassault Systèmes

Building software and agent infrastructure that turns legacy enterprise quality workflows into reliable, inspectable Playwright systems.

  • Architected a modular Playwright and TypeScript framework with reusable API, session, nested-frame, AG Grid, and reporting layers.
  • Integrated Playwright and 3DEXPERIENCE MCP servers for live UI inspection, selector synthesis, and iterative failure repair.
  • Cut a 23-step workflow from 40-50+ minutes to 3-4 minutes while expanding validation checkpoints from 6 to 23.
  • Enabled idempotent reruns with isolated BrowserContexts, persisted workflow IDs, and trace and video artifacts across user roles.
TypeScriptPlaywrightAgent systemsQuality engineering
02

Apr 2026 - Jun 2026

New York, NY

Software Engineer Intern

GTechFin Inc

Designed an end-to-end workflow engine for trend mining, scripting, asset generation, and AI video production.

  • Architected a Python and Typer workflow engine plus MCP service with rerunnable DAG stages, parallel branches, and persisted handoffs.
  • Modeled pipeline state in MySQL with normalized tables, transactions, indexes, foreign keys, and idempotent retries.
  • Benchmarked MongoDB and Elasticsearch, then built a self-hosted cosine index for 500-5K videos with incremental updates.
  • Reduced video production from roughly three hours to twenty minutes across professional and freelance comparisons.
PythonMCPMySQLWorkflow engines
03

Jan 2026 - Present

Irvine, CA

Research Assistant

UCI Relational Cognition Lab

Studying how model families represent relational and virtue concepts through reproducible evaluation and interpretability experiments.

  • Built a PyTorch evaluation harness over 100K+ prompts for Gemma, Qwen, and Llama with standardized prompts, metrics, and runs.
  • Probed layer-wise representations with linear classifiers, cosine similarity, logit lens, and activation patching.
  • Designed random-text injection baselines with repeated trials and variance analysis to separate model signal from prompt noise.
PyTorchEvaluationInterpretabilityResearch
04

Feb 2026 - Apr 2026

Irvine, CA

Research Intern

UCI CS / Xie Lab

Developed a guided AI workflow agent to make bioinformatics analysis more accessible to researchers without programming backgrounds.

  • Designed workflow generation and tool-use assistance around real computational research tasks.
  • Built error handling and recovery support that explains failures and helps users continue an analysis reliably.
LLM agentsTool integrationBioinformatics
05

Sep 2025 - Dec 2025

San Jose, CA

Software Engineer Intern

Fresh Road Inc

Built the real-time backend and operations layer for a multi-tenant AI voice product deployed across thirteen B2B clients.

  • Streamed Twilio audio to Deepgram, language models, and CRM services over WebSockets in a Node.js and Express backend.
  • Orchestrated six BullMQ and Redis queues with retries, backoff, delayed jobs, deduplication, and failed-job recovery.
  • Cut LLM cost 32% and p95 latency 18% while tracing HTTP, WebSocket, queue, model, and CRM paths with correlation IDs and Sentry.
Node.jsWebSocketRedisObservability
06

Jun 2025 - Sep 2025

Guangdong, China

Computer Vision Algorithm Engineer Intern

Ping An Technology

Engineered the data, model, serving, and deployment path for vehicle-damage understanding at production scale.

  • Built a six-stage Python pipeline for multi-view ingestion, quality checks, localization, classification, scoring, and reports.
  • Curated a 2.27M-image training set and fine-tuned Qwen2.5-VL with LoRA, lifting recall 10% and precision 6%.
  • Optimized INT8 serving with dynamic batching, cutting p95 latency 38% and raising throughput 50% for services handling 10K+ daily requests.
PythonComputer visionModel servingKubernetes
07

Jan 2025 - Jun 2025

Irvine, CA

Research Intern

DeepEM Lab, UCI

Built reproducible classical machine-learning experiments for battery cycle-life classification.

  • Built PCA and SVM classifiers on roughly 210K battery records, reaching 87% test accuracy.
  • Standardized preprocessing, cross-validation, and hyperparameter search with pandas, scikit-learn, and MLflow.
Scikit-learnMLflowScientific ML
08

Nov 2022 - Nov 2023

Pittsburgh, PA

Visiting Student Researcher

Carnegie Mellon University

Worked on learned perception for automated cloth folding under Prof. Oliver Kroemer.

  • Developed a neural network for garment keypoint detection to support robotic manipulation of deformable objects.
  • Used large-scale synthetic Blender data to improve generalization and reduce dependence on real-world annotations.
RoboticsComputer visionSynthetic data

Project archive

14 builds
Noted preview

Real-time multimodal transcription and memory

01

Noted

Streams 16kHz PCM audio to Deepgram with sub-300ms latency, combines diarized transcripts with Gemini Vision context, and retrieves memories across sessions with RAG.

Primary stack
Node.js · WebSocket
Focus
Real-time multimodal transcription and memory
Node.jsWebSocketDeepgramGemini
View source
Course Helper preview

Schema-driven academic ETL

02

Course Helper

Converts heterogeneous syllabi into validated, calendar-ready JSON and serves more than one thousand uploads daily with PostgreSQL and Redis.

Primary stack
ETL · PostgreSQL
Focus
Schema-driven academic ETL
ETLPostgreSQLRedisAWS
Live product
LongevityLLM Benchmark preview

Domain LLM evaluation suite

03

LongevityLLM Benchmark

A seven-task evaluation suite with ontology-grounded scoring for genes, alleles, MP IDs, lifespan direction, sex effects, and longevity-pathway reasoning.

Primary stack
Python · Evaluation
Focus
Domain LLM evaluation suite
PythonEvaluationBioinformaticsLLMs
View source
Onchain Narrative preview

Behavior-first Web3 portfolio analysis

04

Onchain Narrative

Turns raw wallet activity into explainable personas, evidence-backed narratives, approval-risk analysis, and chain-aware revoke transactions.

Primary stack
React · Express
Focus
Behavior-first Web3 portfolio analysis
ReactExpressWeb3OpenAI
View source
FocusTrail preview

AI task breakdown and live focus telemetry

05

FocusTrail

A Gemini-powered task orchestration system with a React interface, Server-Sent Events, a macOS focus agent, and a GCP deployment path.

Primary stack
React · Express
Focus
AI task breakdown and live focus telemetry
ReactExpressGeminiGCP
View source
VisaLens preview

ATS and sponsorship analysis extension

06

VisaLens

A Manifest V3 browser extension that identifies sponsorship and work-authorization signals, parses multiple resume profiles, and generates schema-backed fit verdicts.

Primary stack
JavaScript · Manifest V3
Focus
ATS and sponsorship analysis extension
JavaScriptManifest V3GeminiJSON Schema
View source
SousMouse preview

Real-time multimodal cooking copilot

07

SousMouse

A camera- and voice-enabled cooking assistant with FastAPI, Gemini Live, WebSocket streaming, model routing, health checks, and Cloud Run deployment.

Primary stack
FastAPI · Gemini Live
Focus
Real-time multimodal cooking copilot
FastAPIGemini LiveDockerCloud Run
View source
Artified Self preview

Multimodal personal narrative engine

08

Artified Self

A local-first system that protects private context, builds a timeline from desktop signals, and turns a day into a structured, reflective narrative.

Primary stack
FastAPI · Multimodal AI
Focus
Multimodal personal narrative engine
FastAPIMultimodal AIPrivacyData pipelines
View source
Sora Persona Engine preview

Production-ready prompt mining

09

Sora Persona Engine

Analyzes prompt histories and creator patterns to build reusable style guidance and persona-aware short-video generation strategies.

Primary stack
Prompt engineering · Data mining
Focus
Production-ready prompt mining
Prompt engineeringData miningAI orchestration
BetterGI preview

Computer-vision game automation

10

BetterGI

Contributed reliable computer-vision and interaction logic for fishing workflows in an open-source game automation project.

Primary stack
Computer vision · Automation
Focus
Computer-vision game automation
Computer visionAutomationOpen source
View source
Robotic Clothes Folding preview

Learned keypoints for deformable garments

11

Robotic Clothes Folding

Trains perception models on large-scale synthetic Blender data to identify garment keypoints for robotic cloth manipulation.

Primary stack
Robotics · Computer vision
Focus
Learned keypoints for deformable garments
RoboticsComputer visionBlenderSynthetic data
View source
CalGlass preview

AI calorie-recognition glasses

12

CalGlass

A wearable IoT system combining image capture, embedded interaction, server-side inference, and computer vision for food recognition and calorie estimation.

Primary stack
IoT · Computer vision
Focus
AI calorie-recognition glasses
IoTComputer visionEmbedded systems
View source
Epilepsy Detection and Protector preview

Wearable seizure-risk support

13

Epilepsy Detection and Protector

Explores sensor signals and machine learning to identify possible seizure precursors, alert family members, and reduce secondary injury risk.

Primary stack
Machine learning · Sensors
Focus
Wearable seizure-risk support
Machine learningSensorsWearable safety
Smart Anti-Fall Crunch preview

Embedded fall-risk assistive device

14

Smart Anti-Fall Crunch

Combines low-level C, MPU6050 signals, microcontroller control, and GPS modules to detect fall tendencies and trigger protective support.

Primary stack
C · Microcontrollers
Focus
Embedded fall-risk assistive device
CMicrocontrollersMPU6050GPS
View source

Leadership and activities

5 chapters
Chinese Union in Computer Science activity

Event organizer and hackathon judge

Chinese Union in Computer Science

Organized technical events, supported student developer programs, judged hackathons, and mentored teams through project ideas and build decisions.

LeadershipHackathonsMentorship
CSSA activity

Research guidance and student support

CSSA

Created course-selection, lab, professor, and alumni guidance content while supporting information sessions for the student community.

CommunityContentStudent support
Video Game Development Club activity

Unity low-poly multiplayer FPS

Video Game Development Club

Collaborated on gameplay systems and iteration for a multiplayer first-person game in Unity.

UnityC#Multiplayer
VEX Robotics Club activity

Founder and captain

VEX Robotics Club

Founded and led a high-school robotics team, built major robot subsystems, contributed mechanical design, and mentored teammates through competition preparation.

RoboticsLeadershipMechanical design
FIRST Robotics Competition activity

Captain and mentor

FIRST Robotics Competition

Supported mechanism design, team development, CV-based auto-aim, and multi-camera calibration across FRC seasons.

FRCMentorshipComputer vision