Hongcheng
≡
profile photo

Hongcheng Jiang

I am an Applied AI / Machine Learning Engineer working on LLM agent systems and multimodal ML for clinical and scientific imaging. My focus is the reliability half of applied AI—policy-gated tool execution, frozen evaluation suites, trace replay, and failure analysis that names the actual failure mode rather than reporting an average.

I hold a Ph.D. in Electrical and Computer Engineering from the University of Missouri–Kansas City (GPA 4.0), advised by Prof. ZhiQiang Chen.

I am currently a Research Associate (Postdoctoral Researcher) at UMKC, working on software-engineering agents and geometry-conditioned radiographic measurement. Previously I was the sole ML engineer at NextTier, where I built and later open-sourced a bounded agent runtime for radiographic measurement and SQL repair.

I have published 9 papers, including first-author work at IEEE ICIP, IEEE JSTARS, IEEE/CVF CVPR and WACV workshops, and Infrared Physics & Technology, spanning computer vision, image restoration, clinical NLP, and multimodal AI.

Selected work
  › RadMeasure — open-source bounded agent runtime for radiographic measurement and SQL repair. LLM proposes, policy authorizes, deterministic tools execute, verifier decides. On a frozen adversarial suite, policy gating raised task success from 19/36 to 30/36 and blocked every unsafe action. Research prototype only—not a medical device.

Email  /  Google Scholar  /  LinkedIn  /  GitHub

News

Selected Project

RadMeasure — a bounded agent runtime for radiographic measurement and SQL repair. LLM proposes, policy authorizes, deterministic tools execute, verifier decides. On a frozen 36-case adversarial suite, policy gating raised task success from 19/36 to 30/36 and blocked all six unsafe actions; a deterministic rule planner also beat the LLM on a 24-case safety suite — a negative result, and the reason the LLM sits behind deterministic authorization. FastAPI/MCP, async workers, PostgreSQL/MinIO, CI, trace replay.

Research prototype. Not a medical device, and not approved for diagnosis or patient care.

Technical Skills — LLM Agents, MCP, RAG, Agent Evaluation, LoRA, vLLM, PyTorch, FastAPI, PostgreSQL, Docker, Trace Replay, Medical Imaging

Research

My current research is on software-engineering agents and geometry-conditioned radiographic measurement. Earlier work focuses on hyperspectral image processing, image restoration, and vision transformers. Specifically, I have worked on hyperspectral pansharpening using diffusion priors and spectral transformers, thermal and NIR image super-resolution with flexible attention mechanisms, and image colorization with DCT-based networks. My work leverages Transformer architectures, diffusion models, and frequency-domain methods to improve image quality and computational efficiency in remote sensing and medical imaging applications.

Conference Papers:

  1. GeoVIRA: Geometry-Conditioned Visual Interfaces for Radiographic Angle Measurement
    Hongcheng Jiang et al.
    WACV 2027 | under review
  2. SMLP-KAN: Spectral MLP-KAN Diffusion Prior for Hyperspectral Image Restoration
    Hongcheng Jiang et al.
    CVPRW (PBVS) 2026 | paper
  3. THAT: Token-wise High-frequency Augmentation Transformer for Hyperspectral Pansharpening
    H. Jin*, Hongcheng Jiang*, Z. Zhang* et al. (Co-First Author)
    IEEE SMC 2025 | pp. 6627–6634 | paper
  4. Hyperspectral Pansharpening with Transformer-based Spectral Diffusion Priors
    Hongcheng Jiang, ZhiQiang Chen
    WACVW 2025 | paper
  5. Flexible Window-based Self-attention Transformer in Thermal Image Super-Resolution
    Hongcheng Jiang, ZhiQiang Chen
    CVPRW (PBVS) 2024 | paper
  6. DCT-Based Residual Network for NIR Image Colorization
    Hongcheng Jiang, Paras Maharjan, Zhu Li, George York
    IEEE ICIP 2022 | paper
  7. Crucial Data Selection Based on Random Weight Neural Network
    Jie Ji, Hongcheng Jiang, Bin Zhao, Peng Zhai
    IEEE SMC 2015 | paper

Journal Papers:

  1. Transformer-based Diffusion and Spectral Priors Model for Hyperspectral Pansharpening
    Hongcheng Jiang, ZhiQiang Chen
    IEEE JSTARS, vol. 18, pp. 18962–18977, 2025 | paper
  2. Spatial-Frequency Guided Pixel Transformer for NIR-to-RGB Translation
    Hongcheng Jiang, ZhiQiang Chen
    Infrared Physics & Technology, vol. 126, Art. no. 105891, 2025 | paper
  3. Fully-connected LSTM–CRF on Medical Concept Extraction
    Jie Ji, Bairui Chen, Hongcheng Jiang
    International Journal of Machine Learning and Cybernetics, vol. 11, no. 9, pp. 1971–1979, 2020 | paper

Services

Teaching

Graduate Instructor / Teaching Assistant, 13 semesters across Computer Science and Electrical & Computer Engineering at UMKC (2018–2025). Courses included Computer Vision, Algorithms, Discrete Structures, Electronic Circuits, and Computer Design Laboratory.

Resume