Miguel Pineda

From 7 seconds
to under 500ms.

Portrait of Miguel Angel Pineda
MP-2026 · MIAMI, FLPOWER
Open to roles after Dec 2026.Say hi →

Miguel Angel Pineda, software engineer building offline-first distributed systems and applied-ML pipelines. Computer Science senior at FIU, graduating December 2026.

<500ms end-to-end latency•3 internships•3.98 GPA

Where I’ve worked

3 engineering roles•2 countries•2023 – present
01 / 03 · CURRENT ROLEACTIVE
MAY 2025 – PRESENT

AI Systems Researcher

DOJ-Funded De-escalation VR Training GrantFlorida International University • Miami, FL
  • WebRTC
  • Apple Vision Pro
  • Unity · C#
  • Qwen3.5-35B-A3B
  • Whisper

Distributed systems

Co-architected a local-first, fully offline system that keeps two Apple Vision Pro headsets in sync with a central Mac Studio server over WebRTC. I led the migration off legacy WebSockets, which cut end-to-end latency from roughly 7 seconds to under 500ms.

Applied machine learning

Fine-tuned an LLM (Qwen3.5-35B-A3B) on proprietary transcripts to drive the platform’s character dialogue and tool-calling pipeline. I also fine-tuned a Whisper model into a real-time turn-detection classifier, trained on 41.4GB across 270,946 rows.

Test automation

Built two automated stress and integration test systems: dummy WebRTC clients at 100–500 messages a minute, and a Unity Test Framework suite at 30–150 actions a minute. Together they cut pre-demo bug-hunting from 5–10 business days to 1–2.

Ownership

As one of two engineers, with no product manager or written spec in between, I took feedback from FIU Police Department trainers straight into architecture and scope decisions across three-week sprints. I also wrote the process for redeploying the offline system at each demo site.

02 / 03ARCHIVED
MAY – AUG 2024

Data & Backend Engineering Intern

Elicrom Ltd.Guayaquil, Ecuador
  • Python · pandas
  • MySQL
  • Google Maps API
  • GCP

Data engineering

Built ETL pipelines in Python, pandas and SQL to migrate legacy Excel datasets into a production MySQL-based DBMS, part of a migration that cut delivery dispatch time from 7–10 business days to 4–7.

Systems design

Independently built a fastest-route selection system for delivery dispatch, combining Google Maps API route data with live driver input submitted through the company’s employee portal.

Testing & cloud

Wrote SQL and Python regression tests for every new component, per team standard, and ran the company’s website, DBMS and underlying database on Google Cloud Platform.

03 / 03ARCHIVED
MAY – AUG 2023

Back-End Development Intern

Sistagrosa S.A.Guayaquil, Ecuador
  • Spring Boot
  • PostgreSQL
  • REST

Backend optimization

Refactored Spring Boot service layers to resolve database redundancy bugs, cutting API response times by 30% and speeding up PostgreSQL queries on medical inventory records by 25%.

Systems integration

Engineered a REST integration with a third-party shipping provider to replace manual, phone-based coordination, saving 60–90 labor hours a week.

Things I’ve built

De-escalation VR Trainer

DOJ-funded research, FIU•May 2025 – Present

A VR training platform for FIU Police Department trainers, built by two engineers. I co-architected the fully offline system that keeps two Apple Vision Pro headsets in sync with a central Mac Studio over WebRTC.

I led the move off legacy WebSockets, which took end-to-end latency from about 7 seconds to under 500ms. I also fine-tuned Qwen3.5-35B-A3B for character dialogue and tool-calling, and trained a Whisper turn-detection classifier on 41.4GB of data.

VISION PRO A VISION PRO B MAC STUDIO LLM · WHISPER · SYNC WebRTC · offline LAN ~7000ms → <500ms

AI Attendance System

Personal project•Feb – May 2025

A webcam-based facial recognition attendance system in React. It logs each student once per class per day, and keeps inference and every record local in IndexedDB so it runs fully offline.

View the code
● MATCHLOGGED ONCE / CLASS / DAYSTORE · INDEXEDDBNETWORK · NONE NEEDED

Cassandra Product API

Personal project•Java 17 · Spring Boot 3.1.5

A RESTful product management service with a layered controller, service and repository architecture. It models the Cassandra schema and runs custom CQL queries through Spring Data Cassandra.

View the code
CONTROLLER
↓
SERVICE
↓
REPOSITORY · CQL
↓
CASSANDRA

About me

TOOLKIT
  • Java
  • Python
  • TypeScript
  • C# · Unity
  • SQL
  • WebRTC
  • Spring Boot
  • FastAPI
  • REST · JWT
  • ETL
  • React · Node.js
  • RAG
  • LLM fine-tuning
  • Whisper
  • Reranking
  • Evals · recall@k
  • Claude API
  • PostgreSQL · pgvector
  • MySQL
  • Cassandra
  • GCP
  • Docker
  • GitHub Actions
  • pytest

My name is Miguel Angel Pineda.

I’m a Computer Science senior at Florida International University (GPA 3.98, graduating December 2026) with three engineering internships behind me. My coursework spans operating systems, net-centric computing, computer architecture, and data structures and algorithms.

For the past year I’ve been one of two engineers on a DOJ-funded VR training platform, working across WebRTC networking, Unity test automation, and LLM and speech fine-tuning. Before that I wrote Spring Boot services and Python ETL pipelines for companies in Guayaquil, Ecuador, deployed on Google Cloud and Docker.

If you’re working on something that needs to be fast, work offline, or be measured properly, feel free to reach out.