PREDICTIVE MAINTENANCE · MINING FLEETS · ENGINE HEALTH
I build multi-signal monitoring that reads every engine against today’s load and ambient conditions, not a fixed alarm limit. Proven on high-speed engines running 24/7 at sea.
High-speed engines, generators and navigation on vessels at sea, tolerant of intermittent connectivity.
Power plants running around the clock: gensets, running hours, load and temperature under real operating conditions.
Haul trucks and LHDs emit the same ECU signals. The approach layers on top of the OEM telemetry already in place.
Retrospective validation on multi-year histories from three vessels in service: two independent fleets, more than 25 engine-years of operation. Eleven of the twelve developing failures in the record were flagged ahead of the event, with one week to five months of notice. The twelfth was a cooling-water pump that burst with no leading phase in the data; no trend system catches that one. On the second fleet the findings were written down before the failure log arrived. And the system stays quiet: at most one observation worth raising per engine per quarter, with the most severe escalation level practically silent across four years.
A marine engine ignores the weather. A haul truck cannot.
THE PHYSICS
A marine engine rejects its heat into seawater: thermally stable, effectively infinite. A haul truck rejects its heat into the air itself, and every degree of ambient rise spends cooling margin. Going from 15 °C to 40 °C costs about 25 degrees of cooling reserve, so a fixed alarm threshold is not really a threshold. It is a calendar.
THE PAYOFF
The failures most worth catching consume that same margin: dust-fouled radiators and intercoolers, low coolant, tired pumps, slipping fans. In winter the spare margin hides them; the first hot week exposes them all at once, as a wave of alarms that reads like noise. Score each signal against the value it should have today, in today’s conditions, and the climate turns from a source of false alarms into the input that sharpens real ones.
THE FLEET EFFECT
A fleet holds a reference a single machine never has: peers. Every truck in the pit works in the same weather, so when heat moves the whole fleet together, the unit that separates from the pack is separating for its own reasons. And the comparison keeps working in autumn, when falling temperatures pull a faulty unit back under the fixed limit and a threshold would go quiet.
Summer pushes the whole fleet toward the fixed limit at once: that is the false-alarm season. Unit 07 crosses the limit in early summer, and by mid-autumn cooling weather has pulled it back under, so the fixed alarm goes quiet while the fault is still there. Against its peers it never stops being seven degrees adrift.
The faults you most want to catch only show themselves when it's hot.
The season that produces your false alarms is the season that holds your real ones.
The full cycle, not a slice of it.
COLLECTMAPPREDICTACT
Sensor & ECU telemetry: pressures, temperatures, load, running hours. Tolerant of intermittent connectivity.
Time-series at scale: InfluxDB, PostgreSQL, retention and downsampling.
Per-signal health trends against reference behaviour, not raw thresholds.
Time-to-alarm: when each signal will cross its limit, weeks ahead.
Triage dashboards maintenance teams actually work with every day.
2024–present
Integrated software + hardware platform monitoring high-speed engines, generators and navigation on vessels at sea. Continuous acquisition tolerant of intermittent connectivity, alarm processing and prioritisation, dashboards for crew and shore teams.
OUTCOME
RUNNING IN PRODUCTION ON VESSELS AT SEA · USED FOR WARRANTY VALIDATION
2023–present
Personal and client apps, owned end to end: architecture, design, testing, deployment and production maintenance. Published on the App Store and Google Play with international distribution; subscription and payment flows, 2FA, SMS validation and ID verification.
OUTCOME
2M+ CUMULATIVE IMPRESSIONS ACROSS PUBLISHED APPS
2026–present
Per-signal engine trend analysis, fleet-wide condition classification, time-to-alarm projection and prioritisation of interventions. Ingestion of historical manufacturer data for retrospective degradation analysis; LLM-assisted analysis of engine charts.
OUTCOME
FAILURES CAUGHT AS TRENDS, WEEKS BEFORE ALARMS
2019–2025
Earlier career across high-stakes enterprise software: real-time web platforms for digital entertainment, full-stack delivery at Solera, software architecture at Accenture, and high-volume banking and payment gateways at Worldline. Java/Spring microservices, backend architecture, deployment and monitoring.
OUTCOME
THE RELIABILITY DISCIPLINE MACHINE HEALTH IS BUILT ON
Before machine health I built enterprise software: real-time web platforms, full-stack delivery at Solera, software architecture at Accenture, and high-volume payment gateways at Worldline, where a failed transaction is a real problem for a real person. That is where the discipline comes from. Software that is tested, observable, and able to run unattended for months.
Since late 2024 I have been fully focused on machine health: engine telemetry, time-series storage, per-signal trend analysis and time-to-alarm projection for vessels and power plants. The same approach is ready for heavy machinery fleets, mining trucks and LHDs included, and it is brand-agnostic: I work with the ECU signals the equipment already emits (Sandvik, Caterpillar, Komatsu, Epiroc, Volvo) instead of asking anyone to replace hardware.
On the side I design, ship and maintain my own mobile apps, with over two million impressions across the App Store and Google Play. I am based in Madrid and work remote. English C2. I like the problems where the data is messy, the connectivity drops, and someone downstream has to make a maintenance decision from what I put on screen.
Shipped to the stores, end to end.
Personal and client apps, designed, built, published and maintained by me. Every listing below is live; go press the buttons.
2026 · iOS
Peer-to-peer messaging between iPhones over AirDrop and Bluetooth. No internet, no servers, no accounts; built for emergencies and offline situations.
2026 · iOS / Android / Web
Volleyball scoreboard for live match points, sets and statistics.
2026 · iOS / Android
Futsal team and match management: squads, stats, videos, push notifications and full season tracking.
2026 · iOS
Relaxing lawnmower simulator with realistic mechanics and equipment progression.
2026 · iOS
Casual drill game with realistic physics, destructible materials and unlockable tools.
2025 · iOS / Android
Second-degree social network: find people for services through your contacts and theirs. No spam, just trust.
2025 · iOS / Android
Shift calendar for pharmacists: colour-coded shifts, automatic hours and monthly statistics.
2025 · iOS / Android
Join and organise collaborative sporting activities with people nearby.
2025 · iOS
Fitness with a rewards system: burn calories, earn points, cash them in.
2024 · iOS
Facial authentication via ARKit gesture recognition. Biometric security without fingerprints.
2023 · iOS
Create and manage friend groups and events: plans, voting and coordination.
2M+ CUMULATIVE IMPRESSIONSAPP STORE + GOOGLE PLAYINTERNATIONAL DISTRIBUTION
Three steps, each one earning the next.
No new hardware, no downtime, no rip-and-replace: the platform reads the ECU and telemetry signals your machines already emit and layers on top of the OEM system in place. Every step has an exit.
Bring one machine and one month of data. I map signal coverage, baselines and alarm logic, and show you where fixed thresholds are hiding faults or creating noise.
The same test the platform passed at sea, run on your fleet’s history: findings written down first, then compared against the failures your team actually logged. Your own data decides.
A defined subset of the fleet, streaming live, with success criteria agreed before it starts. When the pilot earns it, scale to the rest.
Bring one machine and one month of telemetry.
I’ll review the signal coverage, baselines and alarm logic, then show you where fixed thresholds are hiding faults or creating noise. If your machines already log telemetry, the evidence is already sitting in your files.
REQUEST A TELEMETRY REVIEW