I'm Nick Mino, a rising junior studying Artificial Intelligence at Carnegie Mellon. I build ML systems and product infrastructure, currently as Co-founder, CTO, and ML Engineer at Posematic.
Research: algorithmic stability and post hoc selection in theoretical ML and statistical inference.
open calendly --30minI concentrate in Algorithms and Complexity at CMU's School of Computer Science. My technical work spans computer vision systems, inference infrastructure, product engineering, and theoretical ML research focused on algorithmic stability, post hoc selection, and adaptive generalization guarantees.
Posematic
Working across ML engineering, inference infrastructure, technical architecture, product engineering, and model/data strategy.
Carnegie Mellon University
Researching algorithmic stability and post hoc selection, with a focus on valid generalization guarantees after adaptive model, hypothesis, or parameter selection.
Synthica Journal
Editing computer science submissions for a student-led scientific journal focused on AI and technology.
Carnegie Mellon School of Computer Science
Studying Artificial Intelligence with a concentration in Algorithms and Complexity.
CMU ScottySpark
Posematic won Best Startup after presenting a live prototype demo.
TartanHacks 2026
Team lead for Verdant, a hackathon project focused on the carbon footprint of LLM usage.
HackCMU
Team lead for Ensemble, an audio project using signal-processing-inspired degradation with PyTorch/torchaudio.
Stanford Paleobiology / SEYI
Under Dr. Pedro Monarrez
Worked as a research assistant on paleobiology datasets, including hand annotation of trilobite fossil data and data cleaning, representation, and analysis on mollusk fossil datasets related to California-area records, fossil locality patterns, and missing fossil records.
Non-generative creative tools for artists
Posematic is building non-generative, machine-learning-enhanced creative tools for artists. Our first product converts image or sketch inputs into editable 3D human pose models for fast, detailed reference creation, giving artists control over pose, perspective, and scene construction without replacing their creative process.
Demo video
Company
Posematic
Role
Co-founder, CTO, and ML Engineer
Recognition
Best Startup at CMU ScottySpark; live prototype demo.
ML / AI Systems
Backend / Infrastructure
Product Engineering / Frontend
Product / Leadership
Undergraduate Researcher, Carnegie Mellon University
I work on algorithmic stability and post hoc selection in theoretical ML and statistical inference, with a focus on generalization guarantees for adaptive algorithms. My research studies how to obtain valid generalization guarantees after algorithms adaptively select models, hypotheses, parameters, or procedures using observed data.
When many candidates are evaluated and the best-looking one is selected, the selected candidate's empirical performance can be optimistically biased. This creates a post-selection problem: the data used to choose a candidate also affects the estimate used to evaluate it.
Standard sample splitting avoids post-selection bias by separating the data used for selection from the data used for evaluation, but it can waste data and reduce statistical power. I am studying algorithmic stability as a possible alternative: if the selection procedure is sufficiently insensitive to small dataset perturbations, its selected output may still admit finite-sample generalization guarantees.
This work overlaps with statistical learning theory, but I frame it more broadly as theoretical ML and statistical inference because the project also connects to adaptive data analysis and post-selection inference.
Technical areas
Research Area
Investigating when selection procedures are sufficiently insensitive to small dataset perturbations to support generalization guarantees.
Research Area
Studying generalization guarantees after algorithms adaptively select models, hypotheses, parameters, or procedures from observed data.
Research Area
Characterizing selection-induced bias when the data used to choose a candidate also affects its evaluated performance.
Research Area
Examining how repeated or adaptive use of data affects the reliability of empirical estimates and selected outputs.
Research Area
Connecting finite-sample concentration tools, stability conditions, and alternatives to sample splitting for post-selection guarantees.
TartanHacks 2026 sustainability track winner
Role
Team Lead
Tags
Verdant estimated and reduced the carbon footprint of LLM usage through prompt-complexity analysis, model routing, and context-preserving conversation compression. I led the team and worked on routing, compression, workflow design, product design, and integration.
HackCMU honorable mention audio project
Role
Team Lead
Tags
Ensemble generated vintage-style audio effects using signal-processing-inspired degradation and PyTorch/torchaudio. The project received a HackCMU Honorable Mention.
Experimental chess variant and search engine
Role
Solo Developer
Tags
I built a Python engine for a 3D chess variant as my 15-112 term project. The engine used 8-ply search with custom heuristics, opponent-move assumptions, move ordering, pruning-style shortcuts, and value-threshold short-circuiting to reduce the search space across plies.