My experience connects industrial machine learning, data-driven structural optimization and enterprise software workflows.
01
Professional experience
Dresden, Germany
Industrial measurement technology company · anonymized
Working Student · Data Science & Software Development
Context
Processing, analysis and model delivery for high-frequency industrial sensor time series.
Responsibility
Built an interactive, restart-safe desktop workflow for records of up to 560 million sample points; it supports 1–8 signal pairs with a designed capacity of 4.48 billion sample points. Chunked access and downsampling produced a 70% performance improvement in profiling.
Moved from rule pre-screening to quantized TCN-AE pre-screening, leaving about 5% of each signal for manual correction and improving annotation efficiency by about 60%.
Implemented FCN and CNN-BiLSTM classification, segmentation and anomaly-detection paths: 98% accuracy on a balanced test set and 94%–96% on customer measurement data; quantization reduced model-level CPU inference time by 50% with no apparent loss of accuracy.
Deployed the classification interface to production; segmentation and anomaly detection remained in the test environment. Near-uniform output probabilities were traced to a unit/scale mismatch between customer signals and training data.
Technology
Python
PyTorch
Polars
PyQt
PyQtGraph
TorchScript
Impact
Turned large, difficult-to-inspect recordings into usable labeling and pre-screening workflows, while keeping engineers responsible for final interpretation.
Research Assistant · Data-Driven Structural Optimization
Context
Data-driven design methods for offshore wind substructures and bridge-span layouts under engineering constraints.
Responsibility
Connected synthetic data, predictive models, structural simulation and multi-objective search in an interactive application.
Reformulated bridge layout as sequential decision-making and built Greedy, Beam-DP and reinforcement-learning solvers with reproducible HPC workflows.
Technology
Python
OpenSeesPy
NSGA-II
Flask
DQN
Beam-DP
Slurm
Impact
Made design alternatives and algorithmic trade-offs inspectable for research workflows instead of treating model output as a final answer.
Dresden, Germany
Robotron Datenbank-Software GmbH
Working Student · Energy-Sector Sales Processes
Context
Enterprise ERP data and document workflows for energy-sector sales and billing.
Responsibility
Traced data from user actions through Oracle PL/SQL workflows.
Extended import validation and prepared managed integration tests for document-generation processes.
Technology
Oracle SQL
PL/SQL
ERP data flows
Integration testing
Impact
Translated operational requirements into validated backend data-processing steps within an existing enterprise system.
02
Education
Technische Universität Dresden
German Diplom in Business Informatics · M.Sc. equivalent
Final grade 1.6 (ECTS B) · Advanced Data Analytics; thesis on the design, prototype and evaluation of a fully local RAG system for enterprise product manuals.
Qingdao University
B.A. Finance & B.A. Intercultural German Studies
Dual bachelor’s degrees connecting business foundations, language and intercultural study.
A timeline of systems, constraints and delivery. — Wanjin Li