🎓 CS @ UT Austin | AI & Machine Learning Research | Full Stack Developer | Competitive Programming | Open to Internships

🚗 My portfolio has an Explore Mode — drive a probe around a blueprint of my work: fangedan.github.io

• 🖥 Computer Science Student at UT Austin
• 🔬 Passionate about AI, computer vision, cybersecurity, and competitive programming
• 🎭 Diverse Interests: Theater, piano, web development, and tutoring
• 🌐 Bilingual: English & Chinese

• UTD Machine Learning Research Internship — Prof. Xinfang Jin (2026)

  • Built a Python preprocessing pipeline (preprocess_dream3d.py) to convert DREAM.3D PNG exports into 64×64×64 BMP voxel stacks with measured volume fractions, specific surface areas, and label files — replacing an existing MATLAB workflow
  • Extended the preprocessor with production-ready flags: --multi (batch-processes multiple DREAM.3D stacks sequentially), --tile-xy (spatially tiles large 500×500×500+ volumes into ~49 structures per z-slab instead of downsampling), --dry-run, and --preview; switched downsampling to block-center sampling to keep categorical phase labels exact
  • Validated the full pipeline end-to-end on 101 real DREAM.3D microstructures (first real-data run): generated structures reproduced the real material's Ni connectivity (S=0.905) and pore connectivity (S=0.872) in the OK band on the Yu et al. similarity scale — and Ni connectivity was reproduced without ever being a training target
  • Diagnosed two bugs causing pore-phase collapse in the WGAN-GP (a severed SSA-loss gradient path and a volume-fraction loss computed on softmax instead of argmax), then designed and added a differentiable connectivity loss (3D-convolution isolation penalty + face-hinge percolation term) — fixing pore collapse on synthetic data (similarity 0.48→0.90 and 0.59→0.86, FAIL→OK); a controlled real-data ablation then showed the term overshoots where a phase already percolates, isolating when a connectivity loss helps versus hurts
  • Built 4_CNNCT, a new analysis module measuring phase percolation, active triple-phase-boundary density, tortuosity (taufactor), and Yu et al. distribution-similarity S-values, with a 23-test suite
  • Automated a manual ParaView workflow with 0_PRV/paraview_slice_export.py (pvpython), converting DREAM.3D .vtk volumes into slice-image stacks — shipped with a ground-truth test harness that caught four silent label-corrupting bugs before handoff
  • Wrote a standalone test suite (test_preprocess.py) covering four end-to-end scenarios — resize mode, tile-XY mode, multi-folder mode, and pixel-level phase round-trip — all passing on Windows
  • Built an interactive 3D SOC electrode simulation visualizing the electrochemical process with directional particle flows, deployed via GitHub Pages
  • 🔗 GAN-PH Repository

• UTD Summer Research Internship Program (2024)

  • Studied active speaker detection in Dr. Yapeng Tian's Computer Vision Lab
  • Optimized training techniques (VGG16, transfer learning, weight freezing) to maximize mAP results
  • Built a live in-browser active speaker detection demo — per-face lip tracking correlated with microphone energy, deployed via GitHub Pages
  • 🔗 CVMC Repository

• UTD CAST STEM Bridge Research Lab (2023)

  • Researched hybrid manufacturing & CAD modeling under Dr. Wei Li
  • Designed, coded, and simulated milling paths for medical and automation applications

• President, Merlin Mavens Mentors (CS Tutoring Club)

  • Guided underclassmen through CS concepts and debugging challenges
  • Strengthened mentorship & leadership skills through weekly tutoring sessions

• Computer Science Clubs @ Allen HS

  • 🏆 Competitive Coding (Swift & Python)
  • 🔐 Cybersecurity Club (Competed in AFA CyberPatriot Nationals)
  • 🌐 CS Honor Society

🔹 🗺️ Blueprint Portfolio – This profile, but drivable: a zero-dependency interactive portfolio with a canvas-based Explore Mode, custom drift physics, and an animated percolation background — source on GitHub

🔹 GAN-PH — SOC Electrode Microstructure Generation – Conditional Wasserstein GAN pipeline for generating 3D porous electrode microstructures, validated on real DREAM.3D data with persistent homology and triple-phase-boundary transport analysis, plus an interactive simulation

🔹 CS314 Recursion – Recursive problem-solving in Java, tackling complex algorithmic challenges

🔹 CVMC — Active Speaker Detection – Two-stream audio-visual CNN determining which visible face is speaking, trained on AVA Active Speaker format data — built during the UTD Computer Vision Lab internship and showcased through Freetail Hackers, with a live in-browser demo

🔹 Girlfriend's Website – Silly website I made to ask my girlfriend to be my girlfriend

• 💡 Languages: Java, Swift, Python, HTML/CSS
• 🤖 ML/AI: PyTorch, CNNs, GANs, persistent homology, scikit-learn, OpenCV
• ☁ Certifications: AWS Certified Cloud Practitioner
• 🛠 Technologies: GitHub, Linux, Windows, Cisco, Cybersecurity Tools

• 📧 Email: alin257274@gmail.com
• 🌍 Portfolio: fangedan.github.io ← drive around it
• 💼 LinkedIn