radames.mx

Radames Osvaldo Ramírez Díaz

Lead Software Engineer · Colima, México

Summary

Full-stack engineer specializing in AI-powered applications. Strong backend expertise with solid frontend skills. Builds computer vision, NLP and LLM solutions that ship to production.

Skills

  • Languages: Go, Rust, Python, TypeScript, C#
  • Frontend: React, Flutter, TailwindCSS
  • Backend: Fiber (Go), Axum (Rust), NestJS, .NET Core
  • APIs: REST, SOAP, WebSockets
  • Databases: PostgreSQL, MySQL, MongoDB, Redis, SQL Server
  • AI/ML: PyTorch, Transformers, FAISS, vLLM, YOLO, BERT, Whisper, Coqui TTS, Kokoro
  • Cloud & DevOps: Docker, GitHub Actions, Azure, AWS (EC2, S3, Lambda)

Experience

Gencise — Lead Software Engineer · Jul 2025 – Present

Insurtech building AI underwriting products for commercial insurance (trucking, bars & taverns, property, multi-line), with XPT as the main partner. ~600 PRs across 26 repos so far.

Architecture

  • Designed and built the current ingestion architecture. A central Email Gateway receives every mailbox through Microsoft Graph webhooks and routes each email over RabbitMQ to the product that owns it: age-based priority queues, dead-lettering, effort-based lanes (express/normal/slow) autoscaled on queue depth, alerting, and an admin UI that traces an email end to end. Replaced per-product mailbox polling.
  • Built the shared Python libraries behind it: email consumer, document ingestion (classification, extraction, PDF repair, OCR pre-filter) and the LLM client, used by every product service.

Products

  • Trucking underwriting, email to quote. Led the transportation product for XPT from proof of concept to production: broker submissions arrive by email, LLMs extract ACORD forms, loss runs, MVRs and IFTA recaps, the data is enriched from FMCSA, rated against XPT’s rate revisions, auto-declined when out of appetite, and pushed straight-through as a quote. Increased intake automation and the share of submissions that get quoted. Led its migration from the legacy app to the new platform (IQ Suite).
  • Document extraction. Tamed SOV and loss-run extraction, laid the foundations of ACORD extraction prompting for property insurance, and built part of Bar & Tavern.
  • Model choice. Led the research and A/B tests that made Gemini the main extraction model, which also lowered LLM costs.
  • Email linking. Took follow-up linking further than the first Broker AI version: an LLM tier attaches each follow-up email to the submission it belongs to.
  • Rates. Designed the rating setup that let rate revisions move into the database as self-service with little friction; started that work and handed it off once the foundations were in place.
  • XPT integrations. Submission clearing, straight-through quoting, activity logging and automated broker communication (quote and decline emails) with XPT’s systems.
  • Genie, an internal Slack agent on Claude. Built it to cut the back-and-forth of “who knows how X works”: it answers the team’s product questions from the product data, Jira, Sentry and Slack history, keeps per-product playbooks, proposes Jira/Sentry changes behind a confirm button, and logs and guards production deploys — chores that used to be manual.

SRE and team

  • SRE for the platform. First responder for production fires. Kubernetes on AWS with a shared Helm chart, queue-based autoscaling, Terraform, keyless CI and deploys gated on pending migrations.
  • Team. Mentor, onboard, review code and interview candidates. One of the two people the other devs come to when they’re blocked or want a second opinion — strong opinions, friendly delivery — and part of the architecture decisions.
  • High-priority projects. Get moved onto the most urgent projects, the ones that need to land with the least friction, because I know the business: insurance jargon, underwriting rules and how the company works.
  • Frontend when needed. React + TypeScript across IQ Suite: bar & tavern v2, transportation submission UI, gateway admin, in-browser spreadsheet viewer.

Stack: Python, Celery, Alembic, PostgreSQL, RabbitMQ (Amazon MQ), React, TypeScript, AWS, Kubernetes, Helm, KEDA, Terraform, Claude, Gemini, Sentry.

NebuIA (Distbit) — Full Stack Developer · Mar 2023 – Apr 2025

  • Led full-stack development with Go (Fiber) backends and React frontends: 8 production systems across legal, financial and recruitment sectors.
  • SIGPAD Core (legal case management): Go microservices + React SPA, document processing workflows and case tracking.
  • AiChamba (recruitment): Go backend + React dashboard, Python AI models for candidate screening.
  • Call center app (financial services): real-time WebSockets, voice processing and identity verification.
  • Where I learned machine learning. The CTO’s philosophy was small models that are experts at one task. My first one: a YOLO model that detects the orientation of photographed or scanned documents so Python can straighten them — labeled the dataset, trained overnight, iterated until the accuracy held without overfitting. Also BERT for document classification.
  • Automated KYC. Contributed to a fully automated KYC product, a strong contender in that SaaS market: face recognition with anti-spoofing, and fingerprints read from a phone camera and matched against the voter ID records of Mexico’s National Electoral Institute (INE). The CTO built the foundations and took it to v1; I adjusted and extended it. OpenCV, C++, Python.
  • Microservices infrastructure with Docker and CI/CD for 12+ services.

BiiU — Frontend Developer · Jan 2024 – Jun 2024

  • Reusable React component library (TypeScript + TailwindCSS) shared by 5 product teams.
  • AWS Amplify + AppSync for real-time data sync.
  • GitHub Actions CI/CD: release cycle from 2 weeks to 2 days.

ASIPONA Manzanillo — Full Stack Developer · Mar 2022 – Jun 2023

Mexico’s ports are run by 18 port administrations (ASIPONAs, Administración del Sistema Portuario Nacional) under the Mexican Navy. Manzanillo, the country’s busiest container port, was the pilot for Puerto Inteligente Seguro (PIS): the suite of digital solutions for port operations and the single window of the national port system, built on IMO recommendations and the ISPS Code. PIS was later rolled out to 16 ASIPONAs, and the systems I built went with it. Goals: less friction and more efficient port operations, and auditability for security and against corruption and fraud.

  • First job in the industry. Small team (under 15); I quickly earned independence and usually owned several projects on my own.
  • Started each of these from zero, took it to production, and kept iterating on it:
    • Estado de Hechos: visibility and reporting on container operations at the port.
    • Operational Reports: dashboards and analytics on that operational data.
    • Boletinados: sanctions on port personnel and companies, enforced at physical access control.
    • Embedded Digital Signature: signing microservice for official port documents.
  • Worked directly with stakeholders and end users for feedback, and managed my own priorities.
  • Took a large part in the team’s migration from the legacy monolith to microservices.
  • Backup SRE: covered the Azure infrastructure (databases, storage, deployment pipelines) as the fallback for the senior engineer who owned it.

Stack: C#, .NET, SQL Server, Azure.

Universidad de Colima — Frontend Developer · 2022 – 2023

  • Offline-first PWA (React + service workers) for student attendance across 20 campuses.
  • Federated SSO with C# and SAML 2.0 across 10+ university systems.

Personal project

Susan Doctors — AI virtual assistant for healthcare · 2024 – 2025 · Pilot

Rust, Flutter, Qwen 3 on vLLM (on-prem), Whisper, Coqui TTS.

My first real LLM project. Wrapped Qwen 3 with tool calling and guardrails, ran the inference infrastructure on-prem, and learned the LLMOps I later used at Gencise.

  • WhatsApp assistant for patients (scheduling, questions) + Flutter app for doctors.
  • Rust backend with custom macros for automatic LLM tool calling.
  • Voice-first configuration: doctors set up clinic info, services and staff by talking.
  • Piloted with a medical clinic with good results; wound down after the pilot without funding to keep it going.

Education

  • Universidad de Colima — B.Eng. Internet Engineering Technologies · 2019 – 2023
  • Universidad de Colima — Technical Degree in Programming · 2016 – 2019