01
Perfsy
At Perfsy, I work on ML-driven performance optimization—designing experiments, evaluating model behavior, building data workflows, and carrying what works into the product.
Applied ML / research engineering
I like models that survive the real world.
I’m Daniel, a physicist turned founder and applied ML engineer. I turn ambiguous questions into training and evaluation experiments, then build the systems that carry useful models into production.
MODEL FIELD NOTES / 01
LIVE
FIELD NOTE / DATA
If the data is wrong, nothing downstream can save it.
PICK A STAGE
PHYSICS / ML / SOFTWARE
ASK THE QUESTION+BUILD THE DATA+RUN THE EXPERIMENT+
FIND THE FAILURE+SHIP THE SYSTEM+WATCH IT LEARN+
++
01 / SELECTED WORK
I learned by building the whole thing.
My companies did not come with clean boundaries between research, engineering, and product. I made the calls, wrote the code, and lived with the consequences. That is the kind of ownership I want to keep.
02
Namelor
Namelor taught me what “end to end” actually costs. I handled the product decisions, interface work, backend architecture, and deployment—the useful and unglamorous parts alike.
03
Casabauhaus
Could a machine help an architect think without flattening the creative process? Casabauhaus was my attempt to find out. I built and tested AI-assisted design tools around that tension.
04
lgbm-to-code
I wanted LightGBM models to run where LightGBM could not. So I wrote a small code generator that turns a trained model into plain Python, C++, or JavaScript conditionals, with no runtime dependency on LightGBM.
02 / HOW I WORK
I care about what happens after model.fit().
The dataset is messy. The metric lies. Latency enters the room. I like that part: finding the failure, tightening the loop, and turning a promising model into software people can rely on.
Train the model
Start with a question. Build the dataset, run the experiment, and keep honest notes.
Try to break it
Write the test that could prove the idea wrong. Read the misses, not just the mean.
Build the system
Give the model dependable data, inference, monitoring, and deployment paths.
Watch it work
Put it in front of a real user and stay close enough to see what breaks.
METHODS + TOOLS
PythonPyTorchTensorFlowscikit-learnLightGBM
TypeScriptReactC++ScalaJava
Model evaluationExperiment designNLPComputer Vision
03 / ABOUT
I studied matter at its smallest scales. Then I started companies. Both taught me the same thing: a good answer begins with a better question.
I’m Daniel. I studied condensed matter physics at UC San Diego, then chose the less tidy education of founding software companies.
I have spent the years since moving between equations, code, product decisions, and users. I’m happiest with a hard problem, a fast feedback loop, and teammates who care more about getting it right than looking right.
- Education
- B.S., Condensed Matter Physics
University of California, San Diego - Best fit
- Applied ML
Research Engineering
Model Training & Evaluation - Location
- US work authorized
Open to relocation
04 / CONTACT
Let’s find the part that does not work yet.
If you are training or evaluating models, building the systems around them, or turning research into a product, I would like to hear what is hard.