I'm Emilio. I build agents, automations,
and the tools they live in
I'm a software engineer specializing in automation and applied AI. I design systems where a model handles the part it's good at and deterministic code checks the rest, so people can trust what comes out and keep the workflows they already know.
Mosh
An AI-native digital audio workstation where AI tools live alongside existing VST3 and AU plugins, allowing musicians to augment their workflows with the newest technology rather than being forced to abandon them. A C++ and JUCE audio engine, a React UI, and a Python ML service, all running on one validated command path that the voice agent and the human UI both execute. Every release passes over a thousand deterministic self-tests that run inside the packaged app.
Irresistible
An agentic CPQ engine built to show where a language model belongs in a sales tool. The model has one job: compiling a messy plain-English deal request into a strictly typed DealSpec. Deterministic TypeScript handles everything that has to be correct, pricing tiers and multi-year ramps in integer cents, routing approvals through a parallel graph, and re-validating every alternative before showing it. Thirteen golden tests pin that behavior, and the pricing kernel answers in 0.01 ms at p99.
AskManny
Ask a mixing question and get an answer from the engineers who mixed the records, with timestamped citations back to the exact interview. The pipeline pulls captions from 3,910 videos, splits them into overlapping minute-long segments, attributes each one to its speaker, and indexes more than 99,000 of them in ChromaDB for semantic search. The hosted demo answers through an OpenAI model, and the whole stack can also run fully local through Ollama.
Studies in ML
Generative Spherical Orbifold2024
Escher-style tiled spheres from text prompts, with Nikhil Trivedi, advised by Dr. Crane Chen. Built on Generative Escher Meshes. Ported the L-BFGS solver from MATLAB to PyTorch and CUDA. Shown here is Gingerbread Man. Repo.
CRNN CAPTCHA Breaker2025
CNN + BiLSTM + CTC recognition in PyTorch, reaching 88% full-CAPTCHA and 98.6% character-level accuracy on 10K images, with a failure-mode analysis aimed at stronger CAPTCHA design. Repo.
Background
Education
Northeastern UniversityGraduate coursework in Computer Science (AI/ML focus) · 2024 – 2026
Loyola University New OrleansB.A. Urban & Electronic Music Production · 2020 – 2023
Before software
Years of producing music taught me what a tool feels like when it fights you, and what it feels like when it doesn't.
Earlier on I was a licensed TEDx organizer and event president in high school.