Marcos Medina

Design

Language

EspañolEnglish

Telecommunication Engineer — AI Development & Automation

I engineer intelligent systems that listen, decide and act, from the phone line to the model and back.

Selected work LinkedIn(opens in a new tab)

01 — Profile

Every system I build starts with a signal. A voice on a phone line, the heartbeat of a router, the rhythm of a body in motion, a discharge hidden inside high-voltage equipment. My work is to capture it, understand it and engineer what happens next.

Now
AI Development & Automation at Ausarta
Trained as
Telecommunication Engineer, MSc UPV/EHU
Focus
Voice AI, automation, systems integration

02 — Selected Work

Four systems, four signals.

Voice, infrastructure, the body and the electromagnetic field. Each project starts from a different signal and ends in something that works.

  1. 01 AI Voice Agents Conversational AI × Telephony
  2. 02 Network Operations Platform NOP · Operations × Infrastructure
  3. 03 MarcosHealth Personal project × Health data & AI
  4. 04 Partial Discharge Localization in 3D Master’s thesis × Research
01/04 Conversational AI × Telephony

AI Voice Agents

AI receptionists that answer a company’s phone like its best receptionist would. In production at many companies.

Voice assistants that answer real company phone lines: they understand what the caller needs, put them through to the right person on the company’s phone system, take messages and answer with the company’s own information, at any hour. At the core is an orchestration layer of my own design, between the voice platform and the language model, that supervises the AI in real time.

  • Own orchestration layer
  • Best-in-class STT & TTS
  • Multi-provider inference
  • Yeastar P-Series
  • Email & WhatsApp delivery
  • Blind-judged test bench
View case study
  1. 01

    Ring

    The call comes in through the company’s phone system (a Yeastar P-Series PBX) and the assistant picks up.

  2. 02

    Understand

    Best-in-class speech recognition and synthesis carry the voice. Every turn passes through my own orchestration layer, which supervises the model in real time and fails over between AI providers.

  3. 03

    Validate

    The layer enforces the steps of the process and applies deterministic rules: who the call goes to, what gets confirmed and how a number is read back.

  4. 04

    Act

    The message is read back, confirmed and delivered by email or WhatsApp to the right person, or the call is put through. At any hour.

02/04 NOP · Operations × Infrastructure

Network Operations Platform

The single console the technical team uses to resolve customers’ network and voice incidents.

When a customer is left without internet or without a phone line, every minute counts. NOP brings everything the technical team needs to diagnose and fix it into a single record per customer: routers, 4G backup, phone systems, IP phones and tickets. I have built it from scratch since May 2026, and the whole team uses it every day.

  • MikroTik RouterOS
  • Yeastar
  • Zabbix
  • GLPI
  • Node.js
  • WebSockets
View case study

Seven tools, one record.

Fig. 02 Conceptual interface — not production screenshots

Model-written summary · sample data

“Training load this week sits above your four-week average, while sleep duration has stayed below it for three nights.”

03/04 Personal project × Health data & AI

MarcosHealth

A personal analytics lab for wearable data: trends, context and AI-written readings of recovery and activity.

MarcosHealth gathers health and activity data from wearable devices and turns it into something readable: long-term trends, context around recovery and training, and summaries written by AI models. It describes; it never diagnoses.

  • Wearable data
  • Data visualisation
  • Trend analysis
  • AI models
View case study

Fig. 03 Sample data — no medical interpretation

04/04 Master’s thesis × Research

Partial Discharge Localization in 3D

Locating electrical faults inside high-voltage equipment in three dimensions, with convolutional neural networks and optimization methods.

Partial discharges are tiny electrical breakdowns inside high-voltage insulation and an early sign of failure. They radiate electromagnetic pulses, so their position can be inferred from how those pulses reach a set of sensors. The thesis simulates that propagation with CST Studio, applies convolutional neural networks and optimization methods to estimate the source in 3D, and studies robustness against noise and multipath.

  • CST Studio Suite
  • Convolutional neural networks
  • Optimization methods
  • Signal processing
View case study

03 — Expertise

A working spectrum.

Six bands I operate across, from the physical layer to the product. Hover a band to tune in.

  1. 01

    Voice AI & conversational agents

    Voice agents in production on real company phone lines, built around latency, turn-taking and an orchestration layer that keeps the model in check.

    • Real-time orchestration
    • STT & TTS
    • Multi-provider inference
    • Prompt & tool design
    • Blind evaluation
  2. 02

    Automation & systems integration

    Connecting models, telephony and business systems so a conversation or an event ends in a validated, automated action.

    • APIs & webhooks
    • Business-rule validation
    • Workflow automation
  3. 03

    IP telephony & phone systems

    Voice infrastructure from the inside: phone systems (PBX), IP telephony and the call flows that tie them together.

    • Yeastar P-Series
    • IP PBX
    • VoIP
    • Call flows
  4. 04

    Networks & infrastructure

    Routers, devices and the monitoring that keeps them observable, with an operator’s eye for what actually matters.

    • MikroTik RouterOS
    • Zabbix
    • Fortinet
    • L2TP & VPN
  5. 05

    Product & platform development

    Backends and interfaces that make complex systems usable: operations platforms, consoles for technical teams, product thinking.

    • Node.js
    • Express
    • Frontend
    • Data visualisation
  6. 06

    Signal processing & machine learning

    From electromagnetic simulation to neural networks: research-grade work on extracting position and meaning from signals.

    • CNNs
    • Optimization
    • CST Studio
    • Signal analysis

04 — About

Telecommunication engineer by training, builder by habit.

I’m Marcos Medina, a telecommunication engineer specialising in AI and automation, and I work where infrastructure and intelligence overlap. My background runs through IP telephony, phone systems (PBX), networks and infrastructure; today I design and ship AI systems (conversational voice agents, automations and integrations) together with the product layers that make them usable and observable.

I like problems that span layers: a millisecond budget on a phone line, an outage confirmed on the router before it raises the alarm, a wavefront inside a metal enclosure. Understanding the whole path is usually where the best solution is.

Experience

  1. AI Development & Automation

    Current

    Ausarta

    Conversational voice agents, automation and systems integration, from telephony to product.

  2. Telecommunications & infrastructure

    Hands-on experience with IP telephony, phone systems (PBX), networks, infrastructure and monitoring platforms.

Education

  1. MSc in Telecommunication Engineering

    University of the Basque Country (UPV/EHU)

    Qualifying master’s degree for the regulated profession of Telecommunication Engineer. Thesis on 3D localization of partial discharges.

  2. Telecommunication Engineering

    Undergraduate studies in telecommunication engineering.

Areas of specialisation

  • Voice AI
  • Conversational agents
  • Automation
  • Systems integration
  • IP telephony
  • Phone systems (PBX)
  • Networks
  • Monitoring
  • Product development
  • Signal processing
  • Machine learning

05 — Contact

Open a channel.

For engineering roles, collaborations or a good technical conversation.