Chia En Chang (張嘉恩)
I am an
Specializing in Large Language Models (LLMs), Computer Vision, and Full-Stack Solutions
About Me
As a Computer Science graduate from the University of Arizona, I am an AI Engineer with a strong passion for software development and machine learning. I have hands-on experience in Large Language Model (LLM) development, AI system integration, and advanced image analysis automation. I excel at transforming innovative ideas into effective, AI-driven solutions, from building intelligent systems that integrate language models and computer vision to developing full-stack interactive platforms for complex data analysis and reporting challenges.
Work Experience
AI / Machine Learning Engineer
Anivance AI
- Built GPU-accelerated biomedical image analysis API (WRG) with Cellpose, reducing cell segmentation time by 82% (45s→8s) and enabling same-day experiment results.
- Implemented MCP-style tool orchestration registry with dynamic LLM-based routing across 7 specialized biomedical analysis tools.
- Developed academic report generator (Textara) with RAG system searching 4 databases (PubMed, arXiv, Semantic Scholar, OpenAlex) in parallel.
- Built citation verification engine using DeBERTa-v3-large NLI + SciBERT dual validation, filtering unreliable secondary sources with >98% confidence.
- Architected multi-LLM orchestration layer with hot-swappable backends (GPT-4, Claude, Gemini), implementing dynamic routing based on task complexity.
- Designed neural-symbolic reasoning system with Neo4j knowledge graphs and Logic Tensor Networks, achieving fully explainable AI outputs.
- Built production LLM evaluation infrastructure benchmarking GPT-4, Claude, and Gemini against domain-expert baselines across 5+ biomedical task categories.
- Key Project: Organ-on-Chip Virtual Laboratory — NVIDIA Omniverse Digital Twin (NVIDIA Warp, Isaac Sim, BoTorch, PyTorch, SciPy, USD).
Research Assistant — Adversarial ML
University of Arizona AI Lab
- Developed Universal Adversarial Malware Perturbations (UAMP) using GPT-2, generating adversarial samples that successfully evaded commercial detectors.
- Designed Inverse Reinforcement Learning (IRL) defense against Mosaic Prompt Attacks, predicting and blocking multi-step LLM jailbreak attempts before completion.
- Implemented RLAF framework for automated attack simulation and defense evaluation, reducing manual red-team iteration cycles.
- Validated attack pipelines against VirusTotal API, identifying evasion vulnerabilities across 3 categories of commercial ML-based classifiers.
B.S. in Computer Science
University of Arizona
- Member, University of Arizona AI Lab — Research focus: Adversarial ML, LLM Security.
Technical Skills
AI & Machine Learning
PyTorch, TensorFlow, LLM APIs (OpenAI, Claude, Gemini), RAG, Hugging Face, Reinforcement Learning
NLP & LLM
GPT-2/4, DeBERTa, SciBERT, Transformer, Prompt Engineering, Multi-Agent Systems, LLM Security
Computer Vision
Cellpose (GPU), OpenCV, scikit-image, CLIP, BiomedCLIP, EasyOCR
Simulation & Digital Twin
NVIDIA Warp, Isaac Sim, Omniverse, USD, Lattice Boltzmann Method, SciPy, BoTorch, PhysiCell
Backend & Infrastructure
FastAPI, Flask, WebSocket, Docker, GCP Vertex AI, Neo4j, asyncio
Programming Languages
Python (primary), JavaScript, SQL, Java, C++
Key Projects
Organ-on-Chip Virtual Laboratory — NVIDIA Omniverse Digital Twin
Built an NVIDIA Omniverse digital twin for organ-on-chip virtual drug evaluation, integrating Isaac Sim with a 3-tier GPU-accelerated simulation pipeline.
WRG_API - GPU-Accelerated Biomedical Image Analysis
Built GPU-accelerated biomedical image analysis API with Cellpose, reducing cell segmentation time by 82% and enabling same-day experiment results.
Textara - Academic Report Generator & Citation Engine
Developed academic report generator with RAG system searching multiple databases in parallel, equipped with a citation verification engine.
Neural-Symbolic AI Reasoning System
Designing 5-layer domain-agnostic architecture with pluggable LLM generation layer, targeting explainable AI outputs for clinical audit compliance.
UAMP & RLAF - Adversarial ML Defense
Developed Universal Adversarial Malware Perturbations (UAMP) and RLAF framework for automated attack simulation and defense evaluation.
Contact
I am actively seeking full-time opportunities in software development. I look forward to contributing my skills to your team. Please feel free to get in touch!