Projects

Designer & Sole Developer

A personal operating system that turns the day's incoming clutter — links, slide decks, PDFs, stray to-dos — into searchable, actionable structured assets. It runs on Claude Code and stores everything as plain markdown under git, with no database, no service, and no account. Anything dropped into a Discord channel from a phone is fetched, extracted, classified, templated, and filed — arriving as a pull request for review rather than a silent write.

• Built on one governing constraint: the system's daily writer is an agent, not a person — so every rule in the contributing guide must carry an automated check, and one check verifies that property itself (7 rules, 6 check scripts, one command to run them all)
• Chose markdown + git over Notion or SQLite specifically so every agent write produces a reviewable diff — the precondition for the human merge gate, and the reason full-text search is just grep
• Replaced the official Discord channels plugin with a self-built zero-dependency Python REST polling bridge after diagnosing that WebSocket — and only WebSocket — was blocked on the host network; diagnosis used the gateway HELLO frame as the criterion, since "the socket connected" yields false positives
• Hardened into a always-on service — launchd auto-restart, a file lock so two pollers cannot overwrite each other's watermark, and loud failure on permission denial, because "looks like it is running but is not" is the most expensive failure mode
• Governance patterns distilled from two earlier real projects rather than designed from scratch, converged into a skeleton one person can maintain

Designer & Developer

A floating water-drop character on this site that walks, flies, dances, greets arriving visitors, and answers questions. What began as a static avatar grew into a five-layer system inside a single Astro component — a hand-drawn SVG body with no bitmap assets, a CSS state machine, a JavaScript scheduler with mutual-exclusion guards, a trilingual copy pool, and an on-demand Q&A backend. It is the most personal detail on an otherwise research-toned site.

• Entirely vector: the character is one hand-written 64×64 SVG — body, glassy ears, and every costume are paths and gradients, with no image files anywhere
• Costumes follow one uniform pattern — always present at opacity 0, revealed by a state class — so they cross-fade smoothly and never cause layout shift; adding a new outfit means following the pattern, not inventing one
• Expressions swap whole shapes instead of stretching one: each eye and mouth variant is its own path, so a sleepy squint and a happy arc can each be tuned independently
• Every decorative behaviour is scheduled through explicit mutual-exclusion guards, so the pet cannot start dancing mid-yawn — a constraint that has to be extended deliberately with each new action
• Flies to the navigation bar to point visitors at sections they might otherwise miss, with playful copy written per language rather than translated, since several lines are culture-specific jokes
• Fully honours prefers-reduced-motion on both the JavaScript and CSS sides, and the backend is woken only on an explicit click — never from the idle loop

Solo Founder & Full-Stack Developer

A conversational AI agent that turns kilometer-scale fiber-optic (DAS) sensing data into plain-language bridge-health answers — a verdict, the reasoning, and an exportable engineering report, in one question. Solo-built end-to-end, trilingual (Chinese/English/French) from day one.

• Compresses the full workflow — raw DAS waveform → spectral analysis → modal frequency estimation → health verdict — into a single conversational exchange, computed in seconds to under two minutes
• Produces standard engineering output: PSD spectra, modal frequencies, strain RMS, five-level color-coded alerts, and a one-click archivable Word report
• Built solo end-to-end: frontend, backend, AI agent design, and the underlying DAS signal-processing domain logic
• Trilingual product (Chinese/English/French), switchable in one click — the same i18n design principle later applied to this personal site

DAS Algorithm Expert & Product Lead, Team of 4 — DFW 2026 AI & Startup Competition (Finalist)

A 4-person team competition entry for the DFW 2026 AI & Startup Competition, now a finalist. The team's existing bridge structural-health monitoring dashboard needed a way to answer plain-language questions about live DAS data — I proposed and led embedding a conversational AI assistant directly into that dashboard as one integrated module.

• Advanced to the finals of the DFW 2026 AI & Startup Competition
• Proposed and led the integration of a conversational AI assistant into the team's pre-existing bridge-monitoring dashboard, rather than building a standalone app
• Distinct in scope from MurSense: here the AI assistant is one embedded window within a larger dashboard product built by a 4-person team, not an independent full-stack product built solo

Led development

Developing innovative system for tracking urban mobile sources using DAS technology and edge computing. Focuses on real-time signal interpretation and anomaly detection in smart city context. Currently in implementation phase with Southern University of Science and Technology.

• Integrated edge computing framework with DAS interrogator for real-time processing
• Developed signal processing algorithms for mobile source identification and tracking
• Collaboration with Southern University of Science and Technology — second collaborative project (ongoing)

Co-investigator (Co-first author)

Comprehensive dam safety monitoring project using DAS technology. Comparative analysis of fiber-optic sensing versus traditional seismometers. Paper submitted as co-first author to peer-reviewed journal.

• Deployed DAS system for continuous structural health monitoring of an earthfill dam
• Conducted comparative analysis: DAS vs traditional seismometers for dam monitoring
• Co-first author on paper submitted to Earthquake Science (under review)
• Demonstrated structural reinforcement detection after curtain grouting intervention

Algorithm Lead & System Integration Lead

Comprehensive seismic monitoring system combining AI event detection with EarthX seismic localization technology. Integrated end-to-end pipeline for automatic earthquake detection, location estimation, and mechanism analysis. Deployed in Feidong region for continuous one-year monitoring with 1000+ event detections.

• Led algorithm optimization for EarthX seismic localization system: STA/LTA triggering, PhaseNet auto-picking, real-time localization, and focal mechanism solutions
• Deployed MobileNet-based seismic event detection on datasets ranging from 1.8K to 19K samples, significantly improving precision and recall
• Deployed continuous monitoring system in Feidong region for one year, processing 1000+ seismic events with automated detection and localization
• Integrated end-to-end pipeline: real-time signal processing → AI detection → automatic location → focal mechanism analysis
• Standardized signal preprocessing pipelines, data slicing, and cross-line model evaluation for production deployment

Developer

Developed a multimodal neural network combining vibration, audio (via LSTM), and visual data (via YOLO) for vehicle classification and weight prediction.

• Improved estimation accuracy with signal alignment and feature fusion layers
• Drafted full algorithm documentation

Technical Director

Comprehensive urban safety monitoring platform integrating real-time anomaly detection with multi-subsystem architecture. Combines seismic ambient noise monitoring, AI-driven anomaly event detection across multiple infrastructure scenarios (bridges, tunnels, highways, dams, slopes, airports). Platform deployed and operationalized across multiple cities and diverse application scenarios with real-world validation.

• Designed integrated architecture combining Kafka/TDengine real-time data streaming with AI anomaly detection modules
• Developed subsystems for: seismic background noise velocity change monitoring, shallow subsurface imaging via ambient noise inversion, AI-driven anomaly event detection
• Deployed city-scale pilot implementations covering infrastructure categories: bridges, tunnels, highways, dams, slopes, airports
• Established operational procedures for continuous monitoring and emergency response coordination
• Covered anomaly scenarios: subsurface collapse, slope instability, structural resonance, seismic precursors

Project Director / Led development

Led comprehensive development and verification of city-scale infrastructure safety monitoring system across six infrastructure categories: bridges, tunnels, highways, dams, slopes, and airports. Served as overall project director for government validation phase covering 16+ verification projects with complete end-to-end responsibility from deployment planning to product roadmap development.

• Comprehensive project leadership: equipment deployment planning, project scheduling, data acquisition management across 6 infrastructure categories
• Completed 16+ verification projects in the Shenzhen region with full responsibility for data analysis, reporting, and stakeholder communication
• Coordinated 7 emergency response deployments including high-profile cases: a high-rise building oscillation incident, a railway infrastructure collapse incident investigation, and real-time structural assessment during typhoon events
• Generated 7 granted invention patents directly from verification project outcomes, translating field validation results into IP portfolio
• Established product development roadmap based on field validation results, defining technical specifications and deployment strategies for commercial rollout
• Managed multi-party stakeholders: government agencies, infrastructure owners, construction teams, and regulatory bodies

Algorithm Lead & System Integration Lead

Comprehensive multi-year traffic monitoring and analysis solution providing complete end-to-end system for urban traffic intelligence. Developed from initial research phase through production deployment with real customer projects.

• Led evolution from research prototype to production system: AI-driven traffic signal interpretation from ground vibration data → deep learning pipelines for anomaly detection → commercial deployment
• Developed multi-scenario capabilities: vehicle flow monitoring, traffic signal optimization, overweight vehicle detection, road condition assessment, accident detection
• Built scalable architecture supporting real-time processing of distributed sensor networks across multiple traffic scenarios
• Deployed system across multiple cities with validated customer projects demonstrating operational effectiveness
• Integrated subsystems: vibration-based vehicle classification, acoustic signature analysis, visual confirmation (YOLO), real-time weight estimation

Led development

Led the development of an airport runway seismic-sensing monitoring system. An array of 121 three-component node seismometers captured continuous ground-vibration data, enabling extraction of aircraft take-off and landing signatures for trajectory tracking, landing-impact quantification, and real-time assessment of runway structural safety.

• Built an end-to-end analysis pipeline covering real-time vibration triggering, signal processing, feature extraction, and event classification
• Co-developed software that integrates FFT, wavelet analysis, and AI detection modules
• Secured three granted invention patents
• Deployed at a major hub airport in Guangzhou and a major hub airport in Shenzhen with a 121-node seismometer array
• Processed a 792GB large-scale seismic dataset with AirYolo (YOLOv8-based) detection system

Developer

Developed an innovative method and system to estimate pavement friction coefficients in real time using vibration data generated by moving vehicles.

• Designed a multi-module workflow, including vibration acquisition, denoising, and friction inversion calibrated by vehicle behavior and traffic composition
• Enabled non-contact, low-cost, and high-efficiency evaluation of pavement roughness under icy or worn surface conditions for traffic safety monitoring

Researcher

Applied deep learning methods to systematic research on urban traffic vibration signal interpretation and analysis. Developed end-to-end pipelines for urban traffic monitoring through ground vibration signal processing and interpretation.

• Built deep learning pipelines for anomaly detection and intelligent transportation applications
• Developed comprehensive signal processing and interpretation methodologies for urban traffic vibration monitoring

PhD Researcher

Pioneered the application of Sinkhorn divergence in full waveform inversion (FWI), improving misfit convexity and robustness. Developed and published open-source research code advancing geophysical algorithm implementation.

• Compared various optimal transport–based objectives (W1, W2, GHK, Sinkhorn) in multiple synthetic and realistic inversion scenarios
• Published research implementation code on GitHub, demonstrating advanced Python programming practices and contributing to the open-source scientific computing community

Intern

Designed a misfit function using Kantorovich–Rubinstein (W1) distance via SDMM algorithms to replace L2 in FWI.

• Demonstrated improved global convergence and reduced local minima sensitivity; awarded top thesis score

Researcher

Evaluated VTI-induced errors in micro-seismic hypocenter localization; integrated P-wave and S-wave birefringence analysis.

• Conducted systematic analysis of VTI anisotropy effects on micro-seismic localization accuracy
• Integrated P-wave and S-wave birefringence analysis for improved localization

Undergraduate Researcher

Developed a new method for computing group velocity in VTI media and applied shortest-path algorithms in layered models.

• Developed novel method for group velocity computation in VTI media
• Applied shortest-path algorithms to layered velocity models
• Bachelor thesis: Ray Tracing in Vertically Transversely Isotropic Media

Student team member

Contributed to 3D modeling of a complex magnetite ore body from geophysical (magnetic/electrical) profiles and drillhole data, funded by National College Students' Innovation and Entrepreneurship Training Program grant 201210616004 and NSFC grant 41274130.

• Co-developed a lateral-extension-plus-spatial-interpolation method for 3D modeling of complex ore bodies from 2D geophysical profiles and drillhole data
• Co-authored resulting publication in Science and Technology Innovation Herald, 2013