Experience and portfolio
Engineer First. AI Added.
I’m Paul Sloper, an engineer with over 35 years of hands-on work in defence, air traffic, medical imaging, networks and field service. I am studying for an MSc in Systems Engineering. Since June 2026 I have added AI-assisted software development to that skill set. I set the requirements and the tests, the AI writes the code, and I decide what is correct.
35+Years in engineering
MScSystems Engineering
28Tools on this site, all with simulated data
01 · Specialisms
What I Know Well
Surveillance systemsRadar and ADS-B data, ASTERIX feeds, decoding, replay and tracking.
Radio and voiceSoftware defined radio, airband voice, voice switching and datalink decoding.
NetworksLab networks, failover, passive monitoring and diagnostics.
Systems engineeringRequirements, traceability and the life cycle processes of ISO/IEC/IEEE 15288.
Test and verificationTest procedures, fault injection, defect records and evidence packs.
AI-assisted developmentBriefing models like a contractor, then checking their output against a standard.
02 · Safety and verification
How I Keep It Safe
- Requirements first. Each one is traced to a published source where one exists, and gets a test before any code.
- Right method for each check. Inspection, analysis, demonstration or test, chosen per requirement.
- No invented limits. If a value is unknown, the tool holds a marked placeholder, not a guess.
- Simulated data only. Nothing operational is used or shown.
- Honest status. Each result is reported as pass, fail, not implemented or not verified.
What these tools are not
Limits I State Up Front
- Prototype tools. None is certified, and none is part of a formal safety case.
- Verified by my own tests and reviews, not by an independent body.
- A passing test is evidence, not proof.
03 · Unfamiliar equipment
Meeting Kit for the First Time
- Read before touching. The manual, the interface definition and the published standard come first.
- Watch before acting. Observe it passively, then build a model or simulator so I can learn safely.
- Bench, not live. Try changes on a bench or a lab network, never on something in service.
- Write down what I learn. Each fault and its fix goes in a log, so the next person starts ahead.
- Ask the people who use it. The best fixes start with “that’s not how we actually do it.”
04 · Problems solved
Examples, With Evidence
Does every record arrive?Feed Suite: 196,852 records with 0 gaps, 11 of 11 feeds, 649 tests. Most faults sat at the seam between 2 tested parts. Bench tests on simulated data. Case study
Test without real trafficATC Simulator: 1,137 automated checks with none failed, and a 41-case fault injection procedure.
An alarm that must get throughCrash Alarm Simulator: 2 live paths, 8 written test cases, and placeholders instead of invented limits.
Monitor without touchingASTERIX Sniffer: a passive monitor with no address that sends nothing, tested on a lab network only.
Stop the assistant guessingMy own AI skills make it retrieve and label a source before stating a standard as fact.
Is the decode right?A 34-check decoder self-test and 21 reference test vectors caught a bug that looked fine in the code.
05 · Contact
Get in Touch
Questions about the tools or how they were tested are welcome.