Experience

TIANLANG (JACKIE) CHEN · PERSONAL WEBSITE § 2 · EXPERIENCE

§ 2

6 RESEARCH · 1 INTERNSHIP · 2024–2026

Research

U-Michigan Research

P-2026-01 medical AI in progress

Placeholder figure — medical VLM split-learning pipeline (to be replaced)

P-2026-01 FIG. 014

Scalable Split Learning and Quantization for Vision–Language Models in Medicine

Tianlang (Jackie) Chen

Extending a two-partition multimodal split-learning framework toward multi-server collaborative fine-tuning of medical vision–language models while sensitive data remain localized. The work evaluates low-bit intermediate-representation quantization across medical VQA benchmarks to characterize model-utility, information-preservation, and communication-efficiency trade-offs, and studies privacy leakage under feature-inversion attacks.

↗ GitHub

U-Michigan Research

P-2026-02 regulatory genomics in progress

Placeholder figure — EPCOT foundation-model pipeline (to be replaced)

P-2026-02 FIG. 015

Foundation-Model Fine-Tuning and AI Infrastructure for Regulatory Genomics

Tianlang (Jackie) Chen

Fine-tuning the full EPCOT genomic foundation model on HPAP Type 1 Diabetes snMultiome data to predict gene expression from chromatin accessibility. Designed scalable AI infrastructure spanning memory-efficient data systems, parallel preprocessing and loading, distributed training, and reproducible cluster execution, together with gene-centered, cell-specific, and privacy-conscious evaluation pipelines.

↗ GitHub

U-Michigan Research

P-2026-03 LLM systems in progress

P-2026-03 FIG. 016

LLM-Assisted Incentive Design for Mobility Systems

Tianlang (Jackie) Chen

Engineering schema-aware prompts and strict structured-response validation to simulate heterogeneous traveler decisions with LLMs and convert outputs into rewards for incentive design. Built a scalable inference architecture with batched generation, reusable response pools, and interchangeable serving back ends, then integrated multi-armed bandits and constrained budget allocation with replicated sensitivity and held-out evaluation.

↗ GitHub

U-Michigan Research

P-2026-04 AI platform in progress

P-2026-04 FIG. 017

Full-Stack Development of an AI-Assisted Platform for Scientific Data Analytics

Tianlang (Jackie) Chen

Developed an AI-assisted full-stack platform that transforms heterogeneous experimental data into interactive, validated scientific model-fitting workflows for materials research. Integrated LLM-guided data understanding with structured safeguards and human-in-the-loop verification, and trained a physics-informed neural network to initialize nonlinear fitting within a lightweight inference pipeline combining learned representations, constrained scientific optimization, and uncertainty-aware model comparison.

↗ GitHub

U-Michigan Research

P-2026-05 environmental AI in progress

P-2026-05 FIG. 018

AI-Guided Environmental Sensing and Adaptive Monitoring

Tianlang (Jackie) Chen

Developing AI-guided sensing for sparse, dynamic Great Lakes environmental systems under limited sensor coverage and seasonal variability. Builds reproducible workflows for baseline reconstruction, uncertainty quantification, and visualization, and compares random, grid-based, and uncertainty-aware sensor-placement and adaptive-sampling strategies for downstream deployment.

↗ GitHub

U-Michigan Research

P-2026-06 genomic ML in progress

P-2026-06 FIG. 019

Functional Genomic Representation Learning for Microbial Communities

Tianlang (Jackie) Chen

A composable framework that turns microbial genomes into interpretable, parameter-free functional encodings via proteogenic k-mer tokenization and connects them to classical statistics and large pretrained sequence models. Abundance-weighted community embeddings aggregate per-genome spectra into fixed-length phenotypes, with leakage-safe feature engineering and rigorous statistical validation for high-dimensional, low-sample genomic data.

↗ GitHub

Internship

Mars Asia Internship

§3.04 systems complete

§3.04 FIG. 020

Full-Stack Development — Mars Asia (WeCom)

Tianlang (Jackie) Chen

Contributed to WeCom (Enterprise WeChat) full-stack secondary development using JavaScript, Node.js, and Vue.js for internal communication and workflow tools. Supported API implementation, interface development, data migration, and information retrieval, and built automation workflows that improved operational continuity during system migration.

↗ GitHub

FOL. 014 · EXPERIENCE

7 SPECIMENS · v1.0