Agustín
Kamke

About

Engineer across the physical and digital stack.

Electrical Engineer from Pontificia Universidad Católica de Chile, focused on applied AI, Google Cloud, robotics, sensing and industrial systems. I prototype quickly, validate with users and deploy solutions that have to work outside the lab. Most of what I use day to day I taught myself — most recently by designing and building a VTOL fixed-wing drone from scratch, on my own bench.

Based in

Chile · open to relocating worldwide

Work authorisation

EU passport · German

Languages

Spanish (native) · English C1, TOEFL iBT 97 · German

Primary cloud

Google Cloud Platform

Education

Teaching

• IEE2103 — Signals & SystemsTeaching Assistant
• Programmable Electronic SystemsTeaching Assistant

1. Aug 2025 — present

   SALFACORP

   AI Engineer founder of the AI area

   Industrial AI

   Construction and industrial group · organisation of more than 20,000 workers

   Building and scaling applied AI, cloud and sensing solutions for construction and industrial operations, with a strong focus on Google Cloud Platform adoption.

   • Founded the AI area at the company and promoted adoption across an organisation of more than 20,000 workers.
   • Led cloud-migration workflows to Google Cloud so AI can be embedded into day-to-day operations; this migration path is expected to become the company standard.
   • Built SSOMA Studio, a Veo 3-based video editor deployed on Google Cloud to create safety-communication content using the company's own logos, language and communication style.
   • Created a reusable application template based on self-recursive improvement, where user feedback and usage signals continuously feed the next product iteration — typically within two weeks, with same-day changes for high-priority requests.
   • Scaled TruckScan, a volumetric-scanning and operational-visibility solution that already has several deployed devices across Chile.

   • Google Cloud
   • Veo 3
   • Cloud architecture
   • Product design
   • LiDAR
   • Computer vision

2. Jan — Jul 2025

   UC Center for Astro-Engineering

   Robotics & Radiofrequency Engineer

   Robotics + RF

   Research centre, Pontificia Universidad Católica de Chile · unrelated to the role above

   Developed robotic and AI workflows for drone imagery, autonomous data acquisition and antenna/radar characterisation.

   • Built autonomous acquisition workflows combining robotics, positioning, RF instrumentation and software control.
   • Processed drone imagery and field data for inspection, validation and research applications.
   • Programmed robotic systems for repeatable antenna and radar characterisation.

   • Robotics
   • RF
   • GPS
   • Raspberry Pi
   • Arduino
   • Computer vision

3. 2nd semester 2024

   SEIDOR

   AI & Computer Vision Intern voluntary internship during college

   Enterprise AI

   Voluntary internship completed alongside my university studies, focused on applied AI and computer vision for enterprise use cases.

   • Implemented AI and computer-vision solutions for enterprise use cases.
   • Integrated Gemini, ChatGPT and Claude into controlled business workflows.
   • Built Python data-processing flows for automation, assisted validation and digital-transformation initiatives.

   • Python
   • AI
   • Computer vision
   • Enterprise integration

4. Mar — Aug 2023

   Pontificia Universidad Católica de Chile

   Hardware & Software Developer

   Electronics

   • Designed an educational electronics board for sensors and digital communication.
   • The platform was adopted by the Programmable Electronic Systems course and used by more than 150 students.

   FPGA board in its lab enclosure Sensor and control platform

   • PCB
   • KiCad
   • C / C++
   • Sensors

5. Dec 2022 — Mar 2023

   Komatsu

   Engineering Intern reverse engineering on power electronics

   FPGA + Power

   Recovered undocumented control logic from power-electronics hardware and rebuilt it as a maintainable replacement, from HDL through to a fabricable board.

   • Reverse-engineered the behaviour of a GTO firing-module control card by observing the real hardware, with no documentation of its internal logic available.
   • Reimplemented the logic in Verilog on an Artix-7 FPGA and validated it output by output against the original card, driving both with the same pseudo-random binary sequence.
   • Designed a drop-in replacement PCB in Altium around an Altera MAX V CPLD, with 5 V level shifting, open-collector outputs and its own 3.3 V / 1.8 V rails.

   • Verilog
   • FPGA / CPLD
   • Altium
   • Power electronics

Hover to pause · click any card to jump to the full projectTap any card to jump to the full project

Software

Cloud architecture, applied AI and product delivery — things that ship as a deployment

SALFACORP · AI Engineer, 2025 — present

Cloud migration to Google Cloud

Designed the migration path and cloud architecture required to embed AI into day-to-day operations, with shared infrastructure, centralised services and scalable deployment patterns on Google Cloud.

Cloud migration to Google Cloud — foundation for company-wide AI adoption

DirectionThis cloud-first architecture is intended to become the standard deployment model for new AI applications across the company, replacing isolated prototypes with reusable, governed and scalable infrastructure.

SALFACORP · AI Engineer, 2025 — present

SSOMA Studio · AI video editor on GCP

An internal Veo 3-based editor deployed on Google Cloud to generate worker-safety communication content using SALFACORP's own branding, language and corporate communication style. The platform already supports multiple users.

SSOMA Studio in production — multi-user AI video workflow for safety communication

ImpactTransforms the creation of safety communication from a manual content-production task into a repeatable AI workflow, while preserving corporate visual identity and safety messaging standards.

SALFACORP · reusable product architecture

Self-recursive improvement

A reusable application template so deployed AI products do not remain static after launch: usage, feedback and operational friction become structured inputs for the next improvement cycle.

1. 01User feedbackComments, requests and friction points
2. 02Usage signalsAdoption, behaviour and operational context
3. 03AI analysisPrioritise recurring needs and opportunities
4. 04Improve & deployShip the next version back to users

2 weekstarget cycle to absorb recurring user feedback into the product

Same dayfor high-priority fixes when operationally justified

Continuousfeedback loop after deployment instead of one-off prototyping

MethodThe template combines user feedback, product telemetry and AI-assisted analysis to continuously identify what should improve next — retaining human control over prioritisation and deployment.

Hardware & IoT

Sensing, robotics, FPGA and PCB — things that ship as a physical device and have to survive the field

SALFACORP · AI Engineer, 2025 — present

TruckScan · volumetric scanning deployed in the field

LiDAR and AI-driven field solution for truck and operational scanning. The system already has multiple deployed devices across Chile.

Field deployment Sensor head assembly

System output — volume, weight and material per truck

OutputEvery pass returns load volume, weight and material classification keyed to the truck's plate — a loaded run and an empty run shown side by side above.

ScaleTruckScan combines sensing, computer vision and cloud-connected workflows to generate operational visibility at scale, with several devices already distributed across sites in Chile.

Komatsu · engineering internship, Dec 2022 — Mar 2023

GTO trigger module · replacing an obsolete control card

A GTO thyristor firing module — power-electronics hardware — ran on a control card that had to be replaced, with no documentation of what its logic actually did. I reverse-engineered its behaviour from the surviving hardware, reimplemented the logic on an FPGA, validated it output by output against the real card, and designed a drop-in replacement board around a CPLD — currently laid out and pending a pin remap from the development kit to the final PCB.

Real card and FPGA replica, driven from the same inputs

Vivado block design Replacement PCB, Altium

ValidationBoth cards driven with the same pseudo-random binary sequence and compared on the scope: four of the six outputs matched the original at 100%, PIN18 at 90% and PIN14 at 75%. PIN14 diverges when inputs PIN21 and PIN22 are both high, and PIN18 appears to compare signals rather than emit pulses — neither is reproduced by the current model yet.

StackVerilog on an Artix-7 (Nexys 4) in Vivado · Intel Quartus targeting an Altera MAX V CPLD · Altium Designer · 50 MHz oscillator, 5 V→3.3 V level shifting, open-collector outputs and DC/DC rails at 3.3 V and 1.8 V.

UC Center for Astro-Engineering · Jan — Jul 2025

Autonomous drone RF characterisation

GPS-referenced drone platform and RF measurement setup integrating antennas, signal-chain electronics and embedded control.

Field setup System components

SystemDrone GPS · deck GPS · helix antenna · horn antenna · Raspberry Pi 4 · Valon 5007 · RF amplification / switching · Arduino.

Self-taught · personal project

VTOL fixed-wing drone, built from scratch

A fixed-wing aircraft with vertical take-off and landing, designed and built at home: airframe, propulsion, power stage, wiring and flight electronics. No kit and no course — aerodynamics, motor and battery sizing and flight-control setup learned from documentation and from iterating on the bench until it flew.

Bench build Public showcase

Why it is hereNobody assigned this one. It is the clearest evidence of how I learn: take a system I do not know, read until I can size it, build it, and keep fixing it until it works — the same loop I run at work, only without a brief.

Cloud & systems

Google Cloud Platform, Cloud Storage, Compute Engine, Firebase Hosting, cloud architecture, data integration, synchronisation, remote operations, offline-first systems.

AI & vision

Machine learning, object detection, image processing, anomaly detection, computer vision, generative video (Veo 3), LLM integration (Gemini), AI-assisted workflows.

Robotics & sensing

ROS 2, ROS, LiDAR, RoboSense, Livox, cameras, point clouds, 3D mapping, digital twins, autonomous systems, fixed-wing VTOL UAVs.

Programming & data

Python, C++, C, SQL, Bash, MATLAB, NumPy, Pandas, SciPy, automated reporting and data processing.

Hardware & edge

Raspberry Pi, Jetson, Arduino, industrial sensors, PCB design, KiCad, Altium Designer, FPGA and Verilog (Vivado), CPLD (Intel Quartus), power electronics, embedded computers, RF instrumentation.