Santa Catarina, Mexico

Miguel AngelRomero Licon

Software engineer. Golang, React, cloud, and AI pipelines.

I've spent 12 years shipping production systems, from fintech rails to real-time fleet platforms. Today I help build the AI-assisted royalty parser at Royalytics.

Email me Download CV LinkedIn GitHub
telemetry-ingestincoming vehicle events
  • vehicle.position unit-0142 valid Pub/Sub, Go service, BigQuery ingestedfleet-north
  • vehicle.ignition unit-0087 malformed payload held for inspection heldschema mismatch
  • vehicle.speed unit-0213 valid Pub/Sub, Go service, BigQuery ingestedfleet-south
  • device.heartbeat unit-0031 valid Pub/Sub, Go service, BigQuery ingestedfleet-north
Processed Held for review Illustration, not live data
years shipping production software
12+
Golang microservices designed at Movit
7+
faster platform after a React optimization
60%+
engineers led at Ventup
6

Systems that turn messy inputs into reliable data

Royalytics report parser

Labels, distributors and collecting societies all report music royalties in different layouts (XLSX, CSV, TXT, PDF). Royalytics reads every one of them into a single schema, so creators see one clear picture of what they earned.

The key idea: use AI only when it has to. An LLM drafts the extraction rules for a format it has never seen. After a person validates them, plain rules handle every later report in that format.

Stack. Python, LLM extraction, PostgreSQL, AWS (S3, SQS, SNS), Terraform, React, Node.js.

I work on this with the Royalytics team. The code is private, so I'm happy to walk through the architecture in an interview.

  1. Upload

    The report lands in S3 and a parse job goes onto an SQS queue.

  2. Fingerprint

    The service checks which format the file is.

  3. Choose a path

    Format is known

    A validated config parses it with plain rules. No AI cost.

    Format is new

    An LLM drafts a config for about $0.02. Once confirmed, it is cached.

  4. Normalize and route

    Each row gets a confidence score. Confident rows are stored. The rest go to a person.

  • 11 known formats parse deterministically today, including 4 PDF layouts.
  • Layered Postgres tiers keep raw, cleaned and curated data separate.
  • Hexagonal architecture, strict typing, and layer rules enforced in CI.
  • Progress events flow over SNS. Infrastructure is Terraform on AWS.

  • School attendance and trackingLive

    Student data lived in static PDF manuals, disconnected from real-time attendance. A document-processing pipeline extracts it and feeds a live attendance system.

    Visit escuelahub.com

  • Financial transfers platformFintech, private

    Transfers had to move through Mexico's STP interbank rail while staying in sync with a core banking ledger. I integrated STP with Fineract, the open-source core banking platform, into one transfers system.

    STP, Fineract

  • SAP invoicing integrationERP, private

    Invoices were being re-entered by hand. Using SAP's API, they are now generated straight from the internal ERP.

    SAP API, ERP

Where I've built

  • CodersLinkFeb 2026 to present

    Software Engineeron Royalytics, with AI engineering contributionsCurrent

    • Work across a music royalty analytics platform: React web app, Node.js API and a Python parsing service.
    • Contribute to the AI-assisted report parser: LLM-drafted configs that become cached, deterministic rules.
    • Infrastructure as code with Terraform on AWS (S3, SQS, SNS), plus CI/CD and production monitoring.
  • Movit SolutionsDec 2024 to Feb 2026

    IT Leadfleet management and IoT

    • Designed and implemented 7+ scalable microservices in Golang using Clean and Hexagonal Architecture.
    • Built event-driven distributed systems for real-time data processing.
    • Deployed GCP services (Pub/Sub, BigQuery, Cloud Run, AlloyDB, Looker) with Terraform and Docker.
    • Improved overall platform performance by over 60% by optimizing the React front end.
  • VentupOct 2021 to Dec 2024

    Full Stack Developer and Team Lead

    • Led a team of 6 developers through Agile and Scrum cycles.
    • Built responsive UIs with React and TailwindCSS.
    • Built and containerized services with Node.js, Ruby on Rails and Docker.
    • Integrated GraphQL APIs and set up AWS environments.
  • Galt EnergySep 2019 to Oct 2021

    Full Stack Developer and Team Lead

    • Led a team of 3 engineers, overseeing projects and conducting code reviews.
    • Designed CRM V2 using Node.js, Ruby on Rails, Golang and Vue.js.
    • Designed an internal overdue-accounts management system in Ruby and PostgreSQL.
    • Optimized Laravel (PHP) and Eloquent ORM usage, and managed servers on GCP and AWS for high availability.
  • Early career2014 to 2019
    Four roles in web development
    Ventus TechnologyJan 2019 to Sep 2019
    Developed features and improved legacy systems with PHP, SQL and vanilla JavaScript.
    CarbonodevFeb 2016 to Dec 2018
    Built tailored client solutions with Angular, Vue.js, React, Laravel and Ruby on Rails.
    Captiva Digital TecnologiaDec 2014 to Feb 2016
    Integrated the SAT (Mexico tax authority) RESTful API for stamped electronic invoices, with custom solutions in .NET and Java.
    Business & IT Acceleration MéxicoMar 2014 to Nov 2014
    Optimized SQL Server queries, cutting system load time by 20%, and worked on the "One Map 6" vehicle tracking system.

What I work with

Backend and AI

Golang, Python, Node.js, Ruby on Rails, PHP and Laravel, Elixir, GraphQL, LLM document extraction.

Cloud and delivery

AWS, GCP, Terraform, Docker, GitHub Actions, Pub/Sub, BigQuery, Cloud Run, Looker.

Frontend, data and practice

React, PostgreSQL, TailwindCSS, Vue.js, AlloyDB, Clean and Hexagonal Architecture, event-driven systems, team leadership.

hi@m-romero.dev

I'm open to Senior Full Stack, Tech Lead and AI Engineering roles. Email is the fastest way to reach me.

Download CV LinkedIn GitHub WhatsApp