// research
Methods with structure
I am especially interested in methods that preserve the structure of hard combinatorial and temporal problems while still benefiting from modern data-driven techniques.
lseman@portfolio:~$ cat README.md
|
# profile metadata
roleresearcher · lecturer
stackOR · ML · forecasting
affiliationUFSC · Spacelab
statusbuilding research software
$ echo "▋"
I build optimization algorithms, forecasting models, research software, and interactive teaching artifacts for people working with complex decision systems.
# env vars
PAPERS
--
CITATIONS
--
H-INDEX
--
REPOS
--
# scope
research="Branch-and-price, decomposition, hybrid optimization, and learning-aware decision methods."
teaching="Visual explainers and computational material for optimization and machine learning."
software="Open libraries and experiments that make academic methods reusable."
about.md
I am a researcher and lecturer interested in how rigorous optimization methods and modern learning systems can work together, both as scientific tools and as practical instruments for real decision-making.
/* My work moves between theory, implementation, and explanation. I care about mathematically grounded methods, but also making them operational: software that runs well, interfaces that teach clearly, and systems that help people reason about difficult problems. */
/* Across papers, libraries, and interactive materials, I aim to make advanced topics in optimization, forecasting, and machine learning feel coherent rather than fragmented. */
// teaching
Clarity through interaction
As a lecturer, I value explanations that are visual, tactile, and computational, turning abstract models into things students can inspect, manipulate, and test.
// software
Research that ships
I see software as a research output in its own right: a way to share methods, accelerate experiments, and make academic ideas reusable outside a single paper.
research.log
// research domains
/* My research sits between mathematical rigor and computational experimentation, spanning exact optimization, learning systems, and predictive modeling. */
// domain: OR
Operations.Research
Branch-and-Price, decomposition methods, and scalable exact or hybrid strategies for hard optimization problems.
- Algorithm design for structured decision problems
- Exact methods combined with heuristics
- Implementations built for experimentation and reuse
// domain: ML
Learning.Systems
Learning systems designed with awareness of optimization, combinatorics, and the demands of real decision pipelines.
- Optimization-aware machine learning
- Neural methods for structured decisions
- Graph and sequence models for complex systems
// domain: TS
Forecasting.Models
Deep and statistical approaches for forecasting workflows, multi-step prediction, and interpretable temporal modeling.
- Neural architectures for temporal data
- Attention mechanisms and sequence modeling
- Forecasting systems designed to scale in practice
// wordcloud
Research topic cloud
$
wordcloud2 --source=publications.json
--output=topics.png --mode=diagonal
# generated output
Top Terms
Built from titles, abstracts, and keywords across available publications.
software.json
// selected projects
$ cat software.json | jq '.'
{
"name" : "baldes" ,
"what" : "Modern C++ bucket-graph labeling for vehicle routing problems" ,
"stack" : [ "C++" , "Vehicle Routing" , "OR" ]
} ↗
{
"name" : "foreblocks" ,
"what" : "Modular PyTorch library for time-series forecasting" ,
"stack" : [ "PyTorch" , "Time Series" , "Deep Learning" ]
} ↗
{
"name" : "ipym" ,
"what" : "Fast interior-point methods for linear programming with Python bindings" ,
"stack" : [ "C++" , "Python" , "LP" ]
} ↗
{
"name" : "chomp" ,
"what" : "Modular NLP solver with SQP, interior-point, and DFO-L1 strategies" ,
"stack" : [ "C++" , "NLP" , "SQP" ]
} ↗
repos.json
$ gh repo list lseman --limit 6
/* Selected repositories highlighting current software directions and research infrastructure. */
Loading…papers.log
$ git log --oneline --graph --all ./papers
Loading…
socials.yml
// cat socials.yml
Open to research conversations, teaching collaborations, and software work. If you would like to discuss collaboration, student supervision, invited talks, or open-source projects, these are the best entry points.