Orientation / current public state

About Joseph Maxwell

Joseph Maxwell / materials engineer / technical systems investigator

Current stateFOUNDER & PROGRAMME AUTHOR

I make complex systems legible, test what deserves trust and preserve a way back when the answer changes.

Hi, I'm Joseph Maxwell. I'm a materials engineering master's candidate, writer and the founder of Systemica Engineering. I tend to enter situations that are messy, multi-factor, partially documented or full of plausible explanations, reconstruct what is actually happening, identify the hidden constraints and failure mechanisms, and turn the result into something usable.

Systemica is the largest technical expression of that method, but it is not my entire identity. The same way of thinking runs through my materials work, model investigations, engineering design, writing about complex illness, AI governance and wider systems research.

Personal Method

The person and the way I think

I naturally notice connections between things that are usually discussed separately. I can move from microstructure to software architecture, physiology or organisational failure because I am often looking for the same underlying questions: What state is the system in? What pressures are acting on it? What hidden compensation is keeping it functional? What evidence survives disturbance? What can it recover from? What action is actually justified next?

That breadth is useful, but it creates a communication problem too. The internal map can be much denser than the explanation I am initially able to produce. A large part of my work has therefore become translation: building diagrams, state maps, evidence chains and bounded tests that let somebody else inspect the reasoning rather than simply trust that I can see it.

I am not especially interested in making uncertainty look polished. I would rather preserve the contradiction, show what failed, reduce the claim and identify the smallest next test that could prove or break it.

Engineering & Industrial Experience

Bounded records of materials practice, assessment loss reconstruction, and experimental design.

Materials engineering

My formal base is an integrated MEng in Materials Science and Engineering at Loughborough University. Materials taught me to read visible structure as a record of process: bonding, defects, microstructure, heat, loading and damage constrain what a system can do next.

Defence-aerospace materials assessment environment

During industrial work in a defence-aerospace materials-assessment environment, I worked with microscopy and measurement-led reporting, porosity and lack-of-fusion assessment, grain and additively manufactured titanium analysis, surface-measurement workflows, technical documentation and the reconstruction of a thermal-barrier-coating assessment route after expertise had been lost.

Experimental and modelling work

I have used DOE and ANOVA for process optimisation and built neural-network/model-search work around materials descriptors. An early exercise evaluated all 127 non-empty subsets of seven candidate descriptors. The apparent leading explanation changed when the evaluation route changed. That instability became more important to me than the headline score.

Applied engineering correction

My furnace viewport project also became an important lesson in preserved correction. A later audit exposed that a 350–375 mm scale had been carried where the intended lab-scale envelope was roughly 35–40 mm. Systemica's later bounded rerun did not erase the mistake: it retained the useful cartridge architecture, superseded the invalid thermal interpretation and left the corrected mission on HOLD.

Writing and systems research

I also write across materials, modelling, AI, cognition, complex illness, system dynamics and cosmology. The subjects vary, but the recurring concern is the same: systems can maintain a convincing surface while losing margin, legibility, recovery capacity or evidential integrity underneath.

Health & System Architecture

My health did not "inspire" a startup. It changed the engineering requirements.

I live with a combination of fibromyalgia, Ehlers-Danlos syndrome, AuDHD, dysautonomia/POTS, CPTSD, severe fatigue and gastrointestinal illness. The exact expression varies, but the practical result is that pain, upright tolerance, energy, task initiation, memory access and my ability to translate dense internal thought into external communication can all change without much warning.

My cognitive content does not simply disappear when my body is struggling. More often, access and transmission become unreliable: I may still understand the system while becoming less able to hold every dependency in working memory, move cleanly between tasks or explain the whole map in sequence. Interruption then has a disproportionate cost because reconstructing state consumes the same limited capacity needed to continue the work.

That made several requirements non-negotiable. Work needed to survive interruption. State had to exist outside my head. Decisions needed explicit reasons. Failed routes had to remain recoverable. Another worker or model needed enough context to continue without inventing authority. A bad day could reduce throughput, but it should not silently rewrite what was true yesterday.

Health-to-Design Mapping

Lived Constraint Engineering Requirement Systemica Expression
Variable pain, fatigue and upright capacity Bounded work that can stop safely Explicit mission boundaries, checkpoints and resumable routes
Variable access to working memory and task sequence State outside the operator's head Typed state, source maps, DSTM trajectories and HMA checkpoints
High restart cost after interruption Low-friction re-entry Receipts, lineage, handoff packets and declared next gates
Dense associative thought that can be difficult to serialise Inspectable translation Maplas, diagrams, contracts and evidence chains
Inconsistent human or model output No worker is its own authority Separation of execution, evidence, qualification, authority and accepted memory
Failed attempts are expensive to recreate Preserve negative knowledge Rejected routes, HOLD states, supersession and rollback

AI Assistance & Authority Separation

Why AI entered the process

I became unusually capable with language models because I needed an externalisation bridge. As my energy, memory access and ability to move ideas from internal form into words became less reliable, LLMs helped me keep pace with the amount and density of thought I was trying to preserve.

Repeated use exposed the other side of the problem. A model can be helpful, fluent and wrong. It can lose context, reconstruct a plausible history, flatten uncertainty or behave as if producing an answer gives it authority over the state of a project.

Systemica's response is not to pretend AI was absent. It is to separate assistance from authority. GPT/ChatGPT materially accelerated critique and synthesis. Codex performed substantial implementation, testing, execution support, documentation and evidence organisation. The deterministic audit and governance core can run without a generative model, and no model is permitted to promote evidence or accepted memory merely because it generated a convincing output.

From lived constraint to product

A continuous development sequence from physical materials work to bounded model auditing.

Materials taught me to read history in structure assessment work exposed gaps between output and evidence model rankings moved when the evaluation route changed interrupted continuity made recoverability non-negotiable LLM use exposed drift, context loss and authority confusion Systemica separated computation, evidence, authority and memory GIL became the first bounded commercial audit route

Current Scope & Focus

What I am building now

The immediate product is deliberately narrower than the full research estate. GIL stress-tests whether engineering-model conclusions remain stable under declared changes before a team commits more simulation, experiment, fabrication or procurement budget. It returns an evidence pack, a CONTINUE / REVISE / HOLD decision and the smallest useful next validation step.

The wider estate explores what happens after a conclusion is accepted: how state drifts, how unsupported changes are contained, how memory is promoted, how failed routes are preserved and how a technical programme can recover without inventing its own history.

Evidence & Boundaries

Clear separation of what exists today versus what remains open.

What Exists
  • executable and replayed components
  • locked materials-model pilots
  • deterministic EDK replay and adapter routes
  • bounded governance, memory and recovery demonstrations
  • a public, source-grounded Atlas
  • preserved failures and HOLD routes
What Remains Open
  • independent-source GIL replication
  • external customer pilot
  • production deployment and support
  • scientific or physical validation for the relevant engineering claims
  • independently reproduced whole-Systemica mission
  • calibrated live-LLM drift evidence
Closing Principle

The thread through all of this is not that I have one universal answer. It is that I keep returning to the same practical act: enter an unclear system, reconstruct its state, expose the hidden assumptions, test what survives and make the next justified action visible.

If you are dealing with a model, technical programme or engineering decision whose output looks more certain than its evidence, that is the kind of problem I want to help make legible.