I engineer, test, and maintain Personnel Protection Systems—the controls that prevent accelerator operation unless people are outside restricted areas and every required safety condition has been verified.
Scroll through a short introduction to SLAC, the X-ray laser, and the people who use it.
SLAC is a U.S. Department of Energy national laboratory operated by Stanford University. One of its major user facilities is the Linac Coherent Light Source—LCLS—the world’s first hard X-ray free-electron laser.
A national laboratory · a global user facility
A laser strikes a photocathode and releases a compact bunch of electrons. Radio-frequency accelerator structures then bring that bunch to high energy and shape it for the process ahead.
Light → electrons → accelerationThe bunch passes through a long sequence of alternating magnets called undulators. The magnetic field makes the electrons weave from side to side and emit X-rays. Interaction between the electrons and their own radiation organizes the bunch into microscopic structures, amplifying coherent X-ray light.
Undulators · microbunching · coherent light
LCLS produces extraordinarily bright, ultrashort pulses across soft and hard X-ray regimes. Soft X-rays are especially useful for electronic structure and chemical behavior; more penetrating hard X-rays can resolve atomic structure and dynamics inside materials and biological samples.
Femtosecond—and shorter—views of matterResearchers from universities, laboratories, and industry around the world propose experiments and come to SLAC for beamtime. Their fields include chemistry, biology, materials science, atomic and molecular physics, quantum science, and matter under extreme conditions.
An international research community
The pulses can freeze motion fast enough to assemble molecular movies: chemical bonds forming, proteins changing shape, charge moving through quantum materials, catalysts working, and matter responding to the temperatures and pressures found inside planets and stars.
Structure is only part of the story—the change matters tooFacility context and photography: SLAC National Accelerator Laboratory. Photos by Alberto Gamazo and Jacqueline Ramseyer Orrell.
Before an accelerator beam can enter an experimental area, the control system must know that doors and gates are secured, people are outside restricted areas, warning systems are working, keys and emergency controls are in the correct state, and beam-blocking equipment is ready.
These safeguards are called Personnel Protection Systems. They resemble industrial emergency-shutdown systems, but extend across large scientific facilities and thousands of interconnected devices. If one required condition is lost, the system must prevent or stop beam delivery.
Translate the intended safety behavior into requirements that can be traced through software, hardware, wiring, and field equipment.
Develop redundant PLC logic, electrical interfaces, signal maps, access-control sequences, and fail-safe behavior.
Exercise normal and failure conditions on representative hardware, automate repeatable tests, and investigate every mismatch.
Coordinate installation, commissioning, acceptance testing, release records, and as-built verification across disciplines.
Leave the drawings, specifications, software versions, procedures, and evidence aligned for the next engineer.
When an automated PLC test reports the wrong signals, the failure may live in the expected test behavior, a name or address in an engineering document, PLC memory, an I/O module, network communication, an operator command, or a condition that remains latched until reset.
The useful question is rarely just “what broke?” It is: what was the system supposed to do, what did the software command, what did the hardware receive, and what did the operator interface report?
Make the smallest well-understood change. Test both normal and failure behavior. Leave enough evidence for another engineer to reconstruct the work.
I built a retrieval-augmented generation prototype for searching technical document collections. It combined local Llama models, Hugging Face embeddings, PDF ingestion, semantic retrieval, and experiments with chunking and vector storage. The aim was not an oracle; it was faster discovery grounded in traceable source passages, with security and deployment constraints considered from the beginning.