01
Cluewright
Six families of daily logic puzzles, each offered at six calibrated difficulty levels—with original generation systems, careful progression, and hints that preserve the solve.
Independent practiceCalifornia
Difficult technical work, deliberately scoped.
GYSN LLC is the independent software and machine-learning practice of Gabriel Berger. I build original products and take on selected, bounded technical engagements.
01 / Selected work
Working systems, not just ideas.
Independent products and experiments spanning logic, language, sound, and everyday coordination.
Featured workDesigned and built independently
02
A World That Speaks
A small world whose inhabitants use a consequential language the player must learn through observation, experiment, and speaking back.
Also liveProducts, systems, and instruments
Quiet Done
A task list that keeps your words, closes small loops, and can quietly tell someone when a thing is done—without turning life into project management.
One Long Signal
One continuously composed broadcast: original music generated ahead of the listener, transmitting day and night, never repeating.
Colored Noise
A programmable four-voice noise synthesizer with envelopes, movement, effects, JSON composition, and WAV export—all running in the browser.
02 / Practice
A senior engineer for the difficult part.
I am most useful when a problem is technically demanding and still ambiguous. For example: a question that needs a real answer — feasibility, cost, risk — before a team commits; an LLM feature that shipped without an evaluation story, where quality is anecdotal and regressions are invisible; a system that has to be built from scratch and built well. These are examples, not boundaries; the common thread is difficulty, not a particular domain.
Frame the problem
Turn an unclear technical objective into a tractable design, experiment, or implementation plan.
Build the system
Hands-on development, usually end to end: LLM systems and agents, evaluation pipelines, generation and simulation systems, and complete working prototypes.
De-risk the decision
Test an approach, review an architecture, or establish feasibility before committing a larger team.
Typically starting at 4–12 weeks10–20 hours per weekThrough GYSN LLC
03 / Background
Engineering depth with a mathematician’s habits.
I have spent more than 25 years building software and more than a decade working in machine learning. Before industry, I taught mathematics as an assistant professor. I later became a Staff Software Engineer at Google and a Senior Staff Machine Learning Engineer at Coupang.
I now work independently through GYSN LLC, dividing my time between original products, experimental systems, and selected client engagements with a clear objective and boundary.
MathematicsResearch · teachingSoftware25+ yearsMachine learning10+ yearsIndependent practiceGYSN LLC
04 / Contact
Have a difficult, bounded technical problem?
Send a short description of the problem, the outcome you need, and the approximate timeframe. If it appears to be a good fit, we can discuss how to scope it.
contact@gysn-llc.com