居米研究室

CV

黃俊銘 Chun-Ming Huang

Taipei, Taiwan | jimmy60504@gmail.com

LinkedIn: jimmy60504 | GitHub: jimmy60504 | Blog: jimmylab.seisblue.com | SeisBlue: seisblue.com

Google Scholar: Chun-Ming Huang | ResearchGate: Chun-Ming Huang

Applied AI Engineer | Sensor Time Series | Scientific Software & Infrastructure

Applied AI engineer working with high-throughput multichannel time series, spatiotemporal sensor data, scientific imaging, model diagnosis, and real-time inference. Experience is deepest in seismology, where more than a decade of signal processing developed the diagnostic and systems judgment now applied to broader data-intensive AI problems.

  • Diagnoses model failures from inputs, edge cases, training dynamics, and loss geometry, then tests the mechanism with controlled experiments.
  • Turns research models into replayable, verifiable systems; deployed a real-time earthquake-warning system for internal research and validation on Central Weather Administration servers and now leads its commissioned successor project.
  • Connects station-level time-series inference with cross-station association and location, and reconstructs 2D/3D scientific images from multi-receiver wavefields.

Professional and Research Experience

SeisBlue Co., Ltd. | Taipei, Taiwan

AI Application Engineer / Technical Consultant | 01/2025 – Present

  • Led technical delivery of the Central Weather Administration’s EEW-WebUI project through SeisBlue Co., Ltd. as the contracting entity, extending live Earthworm waveform and warning streams into a React, FastAPI, WebSocket, and Redis Streams system.
  • Scaled Plotly to a GPU-accelerated Deck.gl view for nearly 500 live stations; integrated PhaseNet into the development environment with replay and tests.
  • Designed and shipped Blue’s Reserve (reserve.seisblue.com), a full-stack React/TypeScript/Three.js/Supabase product built through an agentic engineering workflow.

National Taiwan University | Taipei, Taiwan

Doctoral Researcher, AI Systems and Scientific Software | 09/2022 – 12/2024

  • Built sensor-data systems spanning continuous-waveform inference, phase picking, cross-station association, event location, and earthquake early warning.
  • Deployed TT-SAM in real time on Central Weather Administration servers for internal research; it ran for more than six months.
  • Led development of SeisBlue, the lab’s open-source seismic AI data-processing platform, later used by peer-reviewed earthquake-sequence research.
  • Operated and expanded the lab’s research computing environment throughout (see Projects).

Academia Sinica, Institute of Earth Sciences | Taipei, Taiwan

Contract Assistant, Applied AI and Scientific Software | 01/2021 – 04/2023

  • Based at the NTU Structural Seismology Lab for a cross-institution Central Weather Administration AI earthquake-early-warning project.
  • Led a small scientific-software team building earthquake deep-learning applications.
  • Established remote collaboration during COVID: development processes, an internal knowledge base, and GitHub-based project management.
  • Maintained lab servers, research web services, and software environments across the four-year project.

National Central University, Department of Earth Sciences | Taoyuan, Taiwan

Full-time Project Staff, Scientific Software and Deep Learning | 01/2020 – 12/2020

  • Developed deep learning models for seismological applications, including earthquake signal detection and location.
  • Built a machine-learning seismic-attribute workflow for carbonate-reservoir characterization in a commissioned industry study.
  • Processed and interpreted reflection seismic data: preprocessing, velocity analysis, stacking, and migration.
  • Trained new lab members in seismic processing, deep learning, and programming; maintained the lab’s computing infrastructure.

National Central University, Department of Earth Sciences | Taoyuan, Taiwan

Full-time Project Assistant, Reflection Seismic Processing | 10/2015 – 04/2017

  • Led the reflection-seismic processing group; built Echos workflows, scripted job generation, and reusable SOPs for nonstandard processing problems.
  • Reconstructed and interpreted high-resolution 2D/3D subsurface images from marine and land wavefields.
  • Authored and assembled a multi-investigator commissioned report, from field acquisition through interpretation and sequence stratigraphy.
  • Trained five master’s students and two assistants in reflection processing; fielded processing questions from beyond the home lab.

CPC Corporation, Taiwan Exploration & Development Research Institute | Miaoli, Taiwan

Geophysics Intern | 07/2013 – 09/2013

  • Completed an eight-week training program in seismic interpretation and gravity/magnetic processing.
  • Performed horizon tracking, fault interpretation, loop mis-tie checking, and time-structure contour mapping on 2D seismic sections.

Projects

EEW-WebUI | Real-time Earthquake Data and AI Integration System | 01/2025 – Present (demo video: seisblue.com)

  • Extended Earthworm waveform and warning streams into a React, FastAPI, WebSocket, and Redis Streams system for real-time monitoring and model integration.
  • Scaled the waveform view from an early 25-station Plotly demo to an optimized Deck.gl interface, making the near-500-station agency feed visible in one view for the first time.
  • Integrated PyOcto for cross-station event association and PhaseNet for automatic P- and S-wave picking in the development environment; replay, tests, benchmarks, CI, deployment packages, and documentation support verification and handoff.

BlueDisc (SeisBlue Discriminator) | Training-dynamics and Shape-learning Research | 2025 (GitHub: SeisBlue/BlueDisc)

  • Plotted model predictions at every training step to expose how S-wave predictions repeatedly rose toward the detection threshold and were pushed back down.
  • Diagnosed that the model lacked guidance about valid output structure, then added a lightweight discriminator as a shape teacher while keeping the original PhaseNet architecture unchanged.
  • Confirmed the mechanism through loss-landscape analysis and numerical experiments without a neural network; qualifying S-phase detections rose 64% on a 244,242-sample INSTANCE test split under the study’s criteria.
  • Led the diagnosis, method design, and visualization; completed the software and validation jointly with coauthors.

Related writing:「AI 為什麼老是漏看 S 波？」(seisblue.com)

TT-SAM Real-time Pipeline | 03/2020 – 12/2024 (GitHub: SeisBlue/TTSAM_Realtime)

  • Turned an existing Transformer-based early-warning model into a real-time system deployed internally at the Central Weather Administration for research and validation.
  • Pipeline processing stayed under 0.5 seconds in the documented configuration, with at least two predictions per second.

SeisBlue Platform | 04/2018 – 2025 (GitHub: SeisBlue/SeisBlue)

  • Created and led an end-to-end deep-learning seismic platform that converts station-level temporal detections into spatially constrained events through network geometry, travel-time physics, association, and location.
  • First-authored the EGU 2023 platform contribution; coauthored peer-reviewed research using SeisBlue-derived catalogs. Public repository archived 07/2025; development continued internally.
  • After the 2022-09-17 Mw6.5 Chihshang foreshock, detected 6,104 earthquakes and located 1,223 within 39 hours (first-authored Taiwan Geosciences Assembly 2023 abstract).
  • Co-redesigned the associator’s candidate scheduling with a lab collaborator: authored the peak-prioritization pass, reducing pick overlap to support the collaborator’s parallelized event association.

Lab Research Computing Environment | 2013 – 2025

  • Planned, built, and operated the lab’s shared computing environment for about thirteen years: roughly 15 hosts and several NAS systems holding about 60 TB of research data on 10 GbE networking, with about 10 regular users.
  • Introduced the lab’s first GPU computing in 2017 with a personally funded machine; as demand grew, moved to compute/storage separation and later adopted Ansible and Slurm.
  • Operations spanned the full hardware lifecycle, Linux administration, RAID and off-site backup/recovery, pfSense/VPN remote access, and the physical network.
  • Connected the remote fiber-optic (DAS) deployment in Hualien back to the lab: with no direct login to the vendor’s undocumented interrogator host, mapped its ports with nmap and Wireshark and packaged the SSH forwards into a one-click tunnel profile.

Blue’s Reserve | Custom Beverage Platform | 02/2026 – Present (reserve.seisblue.com)

  • Designed and shipped an inquiry-first custom beverage platform (React, TypeScript, Three.js, Supabase); modeled the 3D bottle and can assets in Fusion 360.
  • Delivered through an AI-native workflow: coding agents directed with a persistent knowledge base, system contracts, and acceptance gates.

gg-music | 10/2019 – 11/2019 (GitHub: gg-music/gg-music)

  • Taiwan AI Academy team final project: CycleGAN timbre transfer (piano to guitar) and VGG16 music-genre classification on GTZAN, with Dockerized training environments.
  • Within the two-person team, my contribution focused on mel-spectrogram preprocessing, the U-Net/PatchGAN CycleGAN timbre-transfer subsystem, and the initial repository architecture.

Wireless Geophone Prototype | 2014

  • Self-taught instrument electronics during the 2014 Maker wave and assembled a wireless geophone data-logger prototype: 24-bit ADC, SD-card logging, and GPS timing on Arduino, streaming over an XBee link.
  • Won a bronze medal entering the concept as a two-person team in NCU’s entrepreneurship competition; the design freed station layout from cable routing and showed data return in real time.

Side Projects

Quantum Machine Learning Coursework | 2026 (GitHub: jimmy60504/quantum_computing)

  • Published quantum-computing coursework as engineering projects: Qiskit and PennyLane experiments in regression, classifier comparisons, and a hybrid CNN+QNN, with reproducible environments and MLflow tracking.
  • Built three interactive demo Spaces on Hugging Face (Data-Reuploading, QML-Classifier-Explorer, Hybrid-QNN-Explorer).
  • Wrote a technical article on the bottlenecks observed, including unsuccessful ideas.

LeRobot SO-101 Robot Arm | 2026 (Blog: so101-build-note)

  • Assembled and calibrated an open-source SO-101 arm, 3D-printed its parts, iterated physical fits, and documented the build.
  • Collected a teleoperation dataset and trained an ACT imitation policy; both public on Hugging Face (dataset bochi, model my_policy).

3D Printing and CAD | 2026 (Blog: why-3d-printing-is-worth-trying)

  • Learned FDM printing hands-on: slicer tuning, infill and orientation choices, PLA/PETG selection for functional versus decorative parts, and a self-designed mounting bracket.
  • Learned Fusion 360 to turn physical objects into printable models through iterative fit testing.
  • Published three models on MakerWorld; the replacement keycap has drawn around a hundred downloads and dozens of community prints.

Skills

  • Programming and AI: Python, Bash, Fortran, JavaScript/TypeScript; PyTorch, TensorFlow, scikit-learn, MLflow, SeisBench, CNN, U-Net, cGAN, Transformer models, model diagnosis
  • Scientific Data and Real-time Systems: NumPy, SciPy, ObsPy, Pandas, HDF5, SQL, Airflow, Earthworm, MQTT, FastAPI, WebSocket, Redis Streams; multichannel time series, spatiotemporal workflows, scientific imaging
  • Infrastructure and Agentic Delivery: Linux, Docker, Ansible, Slurm, GitHub Actions CI/CD, pfSense, NAS/RAID, VPN, DNS; coding-agent governance, knowledge bases, system contracts, acceptance criteria
  • CAD and Physical Prototyping: Fusion 360, FDM 3D printing and slicer tuning, Three.js asset integration

Certifications and Training

  • 2023 – 「網站滲透檢測課程」(Web Penetration Testing Course, 39 hours), National Taiwan University Computer and Information Networking Center.
  • 2019 – 「台灣人工智慧學校技術領袖培訓班第五期」(Taiwan AI Academy Technical Professional Program, 16 weeks), certificate #AT082096.
  • 2015 – Paradigm GeoDepth 2D: Depth Velocity Model Building and Depth Imaging, vendor certificate of completion.
  • Languages: Mandarin (native); English – IELTS Academic 7.0 (2016), TOEIC 755 (2015).

Education

National Taiwan University | Doctoral Studies in Geosciences | 09/2022 – Present

Structural Seismology Lab, Taipei, Taiwan

Research focus: AI for seismology, earthquake monitoring, and earthquake early warning.

First-authored preprint from the BlueDisc project: Recovering Sub-threshold S-wave Arrivals in Deep Learning Phase Pickers via Shape-Aware Loss (arXiv:2511.06731)

National Central University | M.S. in Geophysics | 09/2017 – 01/2020

Structural Seismology Lab, Taoyuan, Taiwan

Thesis: An Automatic P Phase Picking Toolbox Using Deep Learning Method「利用深度學習為基礎的 P 波自動挑波套件」(hdl: 11296/72y6b8)

National Central University | B.S. in Earth Science | 09/2010 – 06/2015

Teaching

Reflection and Refraction Seismic Practicum (GP3019)「地球物理探勘實習」 | National Central University | 2016-2021

  • Substantially redesigned and delivered the full practicum: field acquisition, Paradigm Echos processing, and interpretation; public course materials at Blog: gp3019-2.
  • The underlying workflows and SOPs were self-built from the vendor manuals, with scripted automatic job generation.

Introduction to Computing: Graphics and Programming Applications (GP1015)「計概－繪圖與程式應用」 | National Central University | 106-1, 106-2, 107-1

  • Reworked and delivered a four-module course: Linux command line and Git/GitHub, GMT cartography, MATLAB, and Python visualization; public materials and exams at GitHub: sgylab/GP1015.

Graduate Mentorship

  • Mentored four master’s students in seismic AI research and software development.
  • Trained eight master’s students and three assistants in reflection-seismic processing across the NCU and NTU periods.

Field Experience

  • 07/2017 – Longitudinal Valley active-fault reflection line, Chishang, Taitung; acquisition crew.
  • 12/2016 – Dazhuoshui River Mouth / Hoping River alluvial-fan reflection survey, Hualien; acquisition and field quality control.
  • 07/2013 – Ilan geothermal 3D reflection survey, Ilan; acquisition crew (undergraduate).
  • Acquisition methods: co-designed a record-everything “blind shooting” approach for SmartSolo nodal deployments.

Publications

Peer-reviewed and Preprints

  • Huang, C.-M., Chang, L.-H., Chang, I.-H., Lee, A.-S., & Kuo-Chen, H. (2025). Recovering Sub-threshold S-wave Arrivals in Deep Learning Phase Pickers via Shape-Aware Loss. arXiv:2511.06731. First author.
  • Sun, W.-F., Pan, S.-Y., Huang, C.-M., et al. (2024). Deep learning-based earthquake catalog reveals the seismogenic structures of the 2022 Mw 6.9 Chihshang earthquake sequence. Terrestrial, Atmospheric and Oceanic Sciences, 35, 5. Coauthor. DOI: 10.1007/s44195-024-00063-9

Selected Technical Reports

  • 2021-2024 – Central Weather Administration AI earthquake-early-warning report series, parts I-IV. Named author in 2023; contributed analysis and writing in 2021-2022; developed the real-time system and drafted report content in 2024. Reports: 2024, 2023, 2022, 2021.
  • 2020 – 「利用深度學習法來自動挑選地震波到時」(Automatic Picking of Seismic-wave Arrival Times Using Deep Learning). Subproject 3 in「109年度地震資料之分析應用」. Central Weather Bureau, pp. 469-482. Named coauthor; built and open-sourced the SeisNN data pipeline used by the project. Report: MOTC-CWB-109-E-06.pdf
  • 2020 – 「應用多重屬性機器學習於碳酸鹽岩儲集層特性描述」(Application of Multi-Attribute Machine Learning for Characterizing Carbonate Reservoir Properties). Formosa Petrochemical Corporation, MOEA Petroleum Fund (109-F0108), pp. 46-47, 55-59, 78-100. Contributor to the compiled report; authored the NCU-side industry-academia final report「應用機器學習方法於碳酸鹽岩之屬性與儲集層參數識別」. (Confidential; no public file.)
  • 2015-2018 – Reflection-seismic commissioned reporting: authored and assembled a 103-page team report on four profiles totaling approximately 10.748 km for an alluvial-fan survey, and contributed to the four-part Central Geological Survey active-fault near-surface series. Registrations: 2018, 2017, 2016, 2015.

Open Source

  • 2020 – SciPy: a small documentation fix, merged upstream; the contribution process is documented as a public walkthrough (Blog: contribute-open-source).

Invited Talks

  • 2022-12-01 – Invited public talk:《AI 浪潮下的地震資料分析》(Seismic Data Analysis in the AI Era). AI Talk lecture series, 台灣人工智慧協會 (Taiwan AI Association), Taipei.

Conference Presentations

  • 2024 – Oral:「SeisBlue AI 地震資料處理平台開發理念」(SeisBlue AI Earthquake Data Processing Platform Development). Taiwan Geosciences Assembly. ResearchGate
  • 2024 – Oral:「地熱探勘台灣 AI 進行式」(Geothermal Exploration in Taiwan Using AI). Taiwan International Geothermal Conference 2024 (delivered in English). ResearchGate
  • 2023 – Poster: SeisBlue: a deep-learning data processing platform for seismology. EGU General Assembly 2023, EGU23-13927. First author. ResearchGate
  • 2023 – Oral:「池上地震網 AI 近即時地震活動觀測」(Chishang Earthquake Network AI Real-Time Earthquake Activity Observation). Taiwan Geosciences Assembly 2023. First author.

Honors and Awards

  • 11/2014 – 「2014年千里馬盃模擬創業實驗競賽」(NCU Entrepreneurial Competition), Bronze Medal (team EarthMA). Topic: Arduino Wireless Geophone Module.
  • 08/2014 – 「第二屆東方盃全國勘探地球物理大賽」(2nd “BGP Cup” National Exploration Geophysics Competition), First Prize; endorsed by the Society of Exploration Geophysicists (SEG).
  • 03/2012 – 「全國學生音樂比賽」(National Student Competition of Music), Chinese orchestra ensemble division, 4th prize (with the NCU Chinese Music Club).

Extracurricular Activities

  • NTU Activities – Jazz Club member and drummer (2023-2024), with continuing drum lessons since 09/2023; Sanda Club member (2024-2025), with boxing training since 04/2023.
  • NCU Chinese Music Club – President and percussionist (2010-2014); organized the 2012 north-district six-university joint traditional-music camp.
  • Quake Band – Drummer (2012 onward); campus and enterprise shows plus self-organized off-campus concerts.
  • NCU Percussion and Instruction – Wind Band percussionist (2011-2014), including instruction at an elementary-school camp; Guitar Club percussionist (2013-2014), self-taught djembe player, and cajon instructor.