Zhang Sheng · 张升

Projects

Projects

Research and engineering projects spanning medical AI, blockchain security, smart contract analysis, and high-performance computing.

2 featured projects8 earlier projects

Recent research and representative engineering work.

Past Role

ZhiYuDao (CancerDAO) — AI Health Companion for Cancer Patients

During my time at CancerDAO as project lead, I led the development of a WeChat mini-program that turns fragmented medical records into structured, longitudinal health archives, with AI report parsing, indicator-trend tracking, and an AI consultation assistant.

The project used Tencent CloudBase serverless infrastructure (approximately 126 cloud functions and MongoDB), Coze-orchestrated LLM workflows, and blockchain-based SBTs for data ownership and consent auditing.

Under Review

ODBB — Oncology Decision Boundary Benchmark

This first-author study investigates systematic blind spots in frontier LLMs on real oncology decisions. ODBB covers 2,005 decision points from NCCN guidelines and colorectal-cancer cases, scored by a deterministic engine and validated by two oncologists.

Across nine frontier models, even a pooled “super-model” left 42% of items unanswered by all models, revealing a capability boundary concentrated in cross-guideline meta-decisions.

Earlier Projects

Engineering and methodological foundations for my current research.

Blockchain

CancerDAO — Blockchain Ownership Workflow

Completed a patient-centered access-control workflow spanning SBT identity, token-level ACL, on-chain auditing, key management, consent signing, and administrative operations.

Security Measurement

Smart Contract Fraud Detection and Obfuscation Analysis

Designed Obs_tool, a transfer-centric bytecode analysis framework, and analyzed 1.03 million contracts to identify more than 3,000 highly obfuscated samples and characterize their financial risks.

Machine Learning

AI and Blockchain Fraud Detection

Proposed ETH-GBERT, combining graph convolutional networks and BERT for blockchain transaction fraud detection, with a multimodal learning framework and transaction-analysis pipeline.

GPU Computing

Acoustic Beam-Tracing Software GPU Acceleration

Used CUDA to optimize memory access, parallel execution, and hardware utilization in industrial acoustic simulation software, achieving nearly 800× acceleration.

Graphics Systems

GPUPixel Project

Contributed to a rendering framework at iFlytek AI Research Institute, implementing virtual-server rendering, enhanced image output, and EGL adaptation.

Software Engineering

Acoustic Beam-Tracing Software Refactoring

Refactored approximately 7,000 lines of Fortran acoustic simulation software into C++ while preserving functionality and improving maintainability.

Recommendation Systems

E-commerce Recommendation System

Undergraduate thesis project implementing personalized product recommendations from user behavior, search history, and purchase patterns.

Internet of Things

GPS and Sensor Technology for Epidemic Prevention

Integrated GPS and sensor data for epidemic prevention, covering system architecture, frontend interfaces, and backend interaction testing.