Python data systemsMachine learningLLM evaluationStatisticsLow-latency C++

Data systems, ML experiments, and statistical notes with low latency instincts.

I build reliable data systems, machine learning workflows, and low-latency software for questions that deserve careful measurement.

View work Contact Resume

model-quality / loop-024

illustrative telemetry

guided loop livehover a node to inspect

Incoming events

stream visible

Capture operating signals without losing the context that makes a downstream decision useful.

52.4k/min

sample event rate

Role fit

Where I do my best work.

I am strongest in hybrid roles where data science, model behavior, applied statistics, systems thinking, and continuous learning all have to meet in production.

Data science with deployment gravity

Production Data Science

I like turning ambiguous operational questions into measured workflows, usable model outputs, and decisions people can inspect.

Model quality, not leaderboard theater

Model Building & Evaluation

I enjoy building models, evaluation harnesses, and monitoring views that make model behavior easier to compare and trust.

Performance-aware implementation

Low-Latency Data Systems

I am drawn to hot paths, fast feedback loops, C++ performance work, and systems where latency changes the product experience.

Curious, careful, domain-first

Applied Research & Statistics

I like learning new domains, working through statistical assumptions, and translating research questions into practical evidence.

How I work

I treat learning as one of my top goals.

The goal is to ship useful systems, then keep learning from their behavior.

Gather requirements

Start by naming the decision, constraint, audience, and failure mode before touching the implementation.

Learn from SMEs

Use experienced subject matter experts to understand edge cases, vocabulary, and what would make the result useful.

Design the solution

Choose a small architecture, data contract, and evaluation plan that can survive real use instead of only a demo.

Implement

Build the simplest version that proves the shape, then harden the parts that sit on the critical path.

Deploy

Ship with the operational hooks needed to debug inputs, outputs, model behavior, and user-facing latency.

Monitor

Watch the system after launch so drift, latency, and user feedback inform the next requirement pass.

feeds the next pass

Featured research and systems work

Selected case studies and technical notes on ML, statistics, and low-latency systems.

Browse writing

Sep 16, 202640 min read

Hard

High Performance Matrix Kernels

This covers the internals of NVIDIA GPUs and some techniques used to make matrix multiplication go brrr!

GPUPerformance

Read case

Fast Paths in lavaan: A Performance Study preview

May 24, 202618 min read

Hard

Fast Paths in lavaan: A Performance Study

A research-style account of a lavaan performance PR, using matched profiling runs to connect internal fast paths with measured latency and allocation reductions.

RlavaanPerformanceStructural Equation ModelingBenchmarking

Read case

systems

C++

May 7, 202618 min read

Hard

Linear Algebra in C++

A from-scratch C++ guide to dot products, transposes, matrix-vector multiply, cache locality, allocation control, blocking, SIMD-aware loops, and measurement discipline.

C++Linear AlgebraPerformanceMachine LearningLow Latency

Read case