University of Michigan · Computer Science
I’m Jerry Deng, a software engineer working across backend systems, AI agents, and applied machine learning.
Selected systems
Built around measurable constraints: latency, data quality, failure recovery, and verifiable outcomes.
A production-constrained pathology pipeline where failed ablations led to the real bottleneck: annotation policy.
A natural-language PowerPoint editor with typed plans, dry-run previews, explicit confirmation, verification, and rollback.
An always-on asynchronous monitoring system that turns 390+ live data streams into ranked, explainable alerts.
Case study 01
Seven model-side interventions failed to fix a persistent empty-tile false-positive problem. Instead of adding another model trick, I traced the failure backward through the data pipeline and found a concentrated annotation-policy conflict.
Field-tested engineering
Across agent workflows and always-on monitoring, reliability came from explicit system boundaries—not from trusting a model to do everything.
PPT workflow automation · Universal Beijing Resort
Built a natural-language PPT editing agent around typed plans, default dry-runs, confirmation gates, read-back verification, non-target checks, rollback, and structured traces.
Backend monitoring · Independent project
Redesigned an always-on market monitoring service as a staged decision funnel: deterministic signals first, sector ranking next, and model analysis only for the highest-value candidates.
Experience
Summer Research Intern · Computational pathology and ML systems
Data Intelligence Solutions Intern · Browser agents and reporting automation
AI Technology Research Intern · Data pipelines and technical research systems
AI Software Engineering Intern · PostgreSQL, Qdrant, and retrieval architecture
Software Engineering Intern · ECG data fusion and LightGBM modeling
Additional analysis
A privacy-first engine for transcript parsing, requirement auditing, course scoring, and availability-aware planning.
Python · constraints · TypeScript contractsPublic codeIterative multi-factor trading strategies built for the JoinQuant research platform.
Python · pandas · technical indicatorsIndependent research archiveRegression, GLMs, bootstrap inference, and permutation tests on 2024–25 NBA data.
Python · statsmodels · NBA APIStatistical analysisAn end-to-end study of predator–prey population dynamics using ecological time series.
Python · statistical modeling · visualizationEcological modelingStatistical analysis of ratings, reviews, and categorical attributes across 200 restaurants.
Python · scipy · hypothesis testingExploratory analysisAbout
I’m a Computer Science student at the University of Michigan. The problems I like live where software meets imperfect reality: data that disagrees, APIs that fail, workflows that were never designed for automation, and models that need hard boundaries.
I’m currently focused on backend engineering, AI agent reliability, and applied ML systems that can be measured, debugged, and trusted.