Human oversight · operational quality · knowledge architecture · evidence and provenance · controlled artificial-intelligence workflows

I am an Ottawa-based systems builder and operational writer. I turn ambiguous ideas, repeated failures, practical experience, and large bodies of source material into structures that people can inspect, test, operate, maintain, and improve.

My work focuses on the layer between artificial-intelligence capability and human accountability:

• defining narrow roles and authority boundaries;
• turning loose requests into executable work orders;
• creating review gates and acceptance criteria;
• preserving sources, versions, and provenance;
• controlling behavioural state, mode, and verification through external operating documents;
• auditing the evidence, conditions, measurements, and human decision paths behind AI-supported claims;
• documenting failures so the system does not pay for the same lesson repeatedly;
• translating complex internal capability into truthful external language;
• keeping consequential decisions under human control.

I am not presenting myself as a machine-learning engineer or autonomous-agent developer. My strongest work is the operating and documentation layer that makes AI-assisted work understandable, reviewable, recoverable, portable, and useful under pressure.

Available for remote contract work, specialist projects, advisory work, and unusual full-time roles.

Email · Public portfolio · Paranoid People Live Longer

Systems in which artificial intelligence performs bounded labour while a human retains authority, responsibility, approval, and the ability to stop or recover the work.

Typical components include:

• intake records;
• work orders;
• agent or role definitions;
• structured handoffs;
• gate levels;
• review and repair loops;
• run ledgers;
• incident and failure records;
• version and provenance controls;
• explicit stop conditions.

External operating structures for making preferred human-AI working conditions explicit, testable, recoverable, and portable across changing models or sessions.

The work can include:

• state integrity across conversation, task, artifact, source, tool, and completion state;
• mode detection and response-scale control;
• separation of conversational familiarity from verified current state;
• evidence-backed capability and completion claims;
• behavioural drift and contamination analysis;
• calibration and repair logic;
• deliberate boundaries between model capacity, platform constraints, and controllable operating conditions.

This is external behavioural control, not a claim of model-weight modification or hidden access to model internals.

Frameworks for examining the process behind AI-supported conclusions rather than treating a model output, score, ranking, forecast, or apparent model consensus as self-explanatory.

The work can include:

• claim decomposition and consequential definitions;
• evidence provenance and material exclusions;
• model, version, tool, and execution context;
• institutional authority, incentives, and practical leverage;
• quantitative measurement, uncertainty, and generalization;
• meaningful human oversight, contestability, appeal, and remedy;
• lawful evidence access and auditability limits;
• bounded findings that preserve material unknowns.

Connected document systems that turn scattered knowledge into durable operating infrastructure:

• Field Manuals;
• standard operating procedures;
• operator cards;
• Systems Papers;
• glossaries;
• decision trees;
• checklists;
• review standards;
• source maps;
• status and risk vocabularies;
• maintenance and revision rules.

Processes that distinguish a platform saying “complete” from work that is actually correct, attached, usable, current, and ready for the next step.

This includes:

• acceptance criteria;
• model-output evaluation;
• source and claim checks;
• wrong-file and wrong-version prevention;
• duplicate and attachment checks;
• rollback and recovery procedures;
• evidence-backed status reporting;
• separating blocking defects from optional improvements.

Methods for recovering useful operating logic from old documents, failed projects, procedural residue, and large archives, then explaining the real capability without inventing credentials or pretending design work is enterprise deployment.

A public proof-of-work series that explains built systems without publishing the complete private operating package.

Documents an external behavioural-control architecture developed to preserve useful human-AI working conditions across model changes.

The paper covers:

• the distinction between model capacity, platform constraints, and controllable operating conditions;
• behaviour as more than voice or persona;
• character and truth as paired requirements;
• state integrity across conversation, task, artifact, source, tool, and completion state;
• mode and response-scale control;
• evidence-backed claims about memory, tools, progress, and completion;
• behavioural drift and contamination;
• high-level calibration and recovery logic;
• explicit human authority and known limitations.

The public edition deliberately withholds raw calibration transcripts, private examples, exact activation and reset blocks, internal control Bibles, the complete reusable test suite, detailed implementation sequences, and consulting-delivery machinery.

Demonstrates: AI workflow governance, behavioural architecture, state integrity, calibration thinking, evidence discipline, failure-to-control translation, and the ability to make a private operating system publicly legible without giving away the complete engine.

Does not claim: modification of model weights, hidden model access, restoration of unavailable capabilities, or a jailbreak.

Documents a private AI Employment Radar built to turn a noisy AI job search into a constraint-first employment-intelligence system.

The system was developed through sequential prototypes, beginning with a deliberately mixed batch of 30 live job listings. Deep reading exposed the difference between a role that looks right conceptually and one that is actually usable, so feasibility was moved ahead of nuanced ranking. The final public architecture also separates evidence from aspiration, learns vocabulary from near misses, and keeps consequential application decisions under human control.

Demonstrates: constraint-first decision architecture, AI-assisted evaluation, human oversight, evidence-backed capability translation, sequential prototyping, failure-to-infrastructure learning, knowledge and vocabulary operations, claims discipline, and public/private boundary design.

The public edition deliberately withholds exact scoring mechanics, personal feasibility constraints, candidate-specific evidence, detailed search configuration, named test records, and the private operating Bible.

Documents a private audit framework built to examine the evidence, model context, institutional controls, measurements, human decision paths, and public presentation behind AI-supported claims.

The framework treats auditability itself as part of accountability. It distinguishes evidence access from evidence quality, formal authority from practical leverage, a human reviewer from meaningful human control, and a numerical result from the broader claim made from that result. Material unknowns remain visible rather than being converted into assumptions.

Demonstrates: AI evaluation and claims auditing, evidence and provenance architecture, human oversight and contestability analysis, institutional power and incentive mapping, quantitative-claim discipline, lawful evidence-access design, bounded findings, and public/private system-boundary design.

The public edition deliberately withholds internal agent specifications, detailed prompts, recursive routing logic, branch and stopping mechanics, field cards, ledger design, validation records, and the complete private operating package.

A complete public reference architecture for twelve narrow artificial-intelligence agents.

The repository includes:

• a universal six-agent control loop;
• an optional five-agent collection-management branch;
• a cross-cutting voice and style router;
• defined inputs, outputs, handoffs, and stop conditions;
• evidence and quality gates;
• gate levels for internal work, review, action, and blocked actions;
• shared schemas, templates, status codes, and risk codes;
• a local validation utility;
• unit tests and repository checks;
• fictional examples and a documented sanitization boundary.

Demonstrates: AI workflow governance, human oversight, role separation, documentation architecture, quality assurance, provenance, testing, controlled system change, and public-safe conversion of a private method.

A small working Python engine that repeatedly mutates symbolic faces through four bounded cycles.

The repository includes executable standard-library code, deterministic and finite run options, bounded state, tests, command-line documentation, and an implementation note that accurately describes what the system does and does not do.

Demonstrates: prototyping, bounded state, deterministic testing, technical explanation, scope control, and the ability to turn an unusual conceptual idea into a working artifact.

A public shelf of substantial finished work in artificial intelligence, cognition, recipe-production methodology, operational systems, systems proof, speculative futures, and experimental literature.

The collection includes:

• Rebuilding the Room - Systems Paper 001;
• From Job Search to Employment Intelligence - Systems Paper 002;
• AI Claims Audit Stack - Systems Paper 003;
• The Dark Cognitive Manual;
• Sixty AI Futures;
• The Recipe Writing Production Guide;
• A Zoo at the End of Winter;
• Character Study.

Demonstrates: systems explanation, long-form analysis, Field Manual construction, subject-matter translation, AI-evaluation architecture, scenario thinking, distinct voice systems, literary architecture, and completion of substantial public artifacts.

The main public index connecting released systems, Systems Papers, documents, creative work, professional proof, the PPLL website, and the controlled boundary between public artifacts and the private production archive.

The public repositories show only part of the larger working practice. Additional inspectable or sanitizable work includes:

• a twelve-agent private operating framework with intake, work orders, review, versioning, and run ledgers;
• a private behavioural-reconstruction control stack used to convert repeated model failures and useful working conditions into explicit operating controls;
• a private employment-intelligence system built around feasibility gating, evidence matching, deep reading, and human-approved action;
• a private AI-claims audit stack built around evidence access, provenance, model context, measurement, oversight, and bounded conclusions;
• a multi-document voice and identity control stack;
• read-only first-pass file inventory and recovery logic;
• digital-product identity, attachment, version, test-order, and rollback procedures;
• classification and preservation rules for a 486-piece physical art archive;
• more than one hundred products and related assets managed through a creative and commercial operating system;
• coordinated doctrine covering strategic option preservation, human–machine authority, provenance-led research, value recovery, and evidence-backed capability translation.

Private source documents, raw calibration material, personal records, customer information, production Bibles, and unreleased systems are not posted publicly.

• Artificial-intelligence workflow governance
• Human-in-the-loop and human-oversight design
• AI evaluation and claims auditing
• Evidence-access and auditability analysis
• External behavioural-control architecture
• Constraint-first AI-assisted evaluation
• State and mode discipline for long-running AI-assisted work
• Behavioural drift, calibration, and recovery logic
• Model-output evaluation and quality assurance
• Documentation and knowledge architecture
• Systems Papers and public-safe architecture explanation
• Source tracking, provenance, and evidence grading
• Quantitative-claim and generalization discipline
• Evidence-backed capability translation
• Standard operating procedures and Field Manuals
• Process mapping and handoff design
• Failure analysis and incident learning
• Version control and controlled change
• Capability translation and claims discipline
• Accessibility-aware, low-friction workflow design
• Technical and plain-language writing
• Operational research and synthesis
• Creative systems and conceptual prototypes

I work best when the problem is real but not yet cleanly named.

Examples:

• a team is using AI but nobody can explain what was actually checked;
• an AI-supported conclusion sounds objective but nobody can reconstruct the evidence, model conditions, measurement, or human decision path behind it;
• a model change has damaged a working process and the team can describe the frustration but not the behavioural mechanism;
• a retrieval system produces plenty of nominal matches but few decisions worth acting on;
• knowledge lives in chats, screenshots, scattered files, and people’s memories;
• an agent or workflow has unclear responsibility and approval boundaries;
• the same mistake keeps happening without becoming a system improvement;
• an old, fragmented, or abandoned body of material contains useful operating logic;
• documentation technically exists but cannot be followed under real conditions;
• a complex capability is valuable internally but sounds strange or inflated when described publicly;
• a private system needs to become inspectable proof without exposing the full operating method;
• a workflow assumes perfect stamina, easy mouse use, and unlimited manual repair.

1. Define the actual job before executing it.
2. Build the container before producing volume.
3. Preserve source material and earlier versions.
4. Separate evidence, inference, recommendation, and speculation.
5. Make roles, inputs, outputs, gates, and stop conditions explicit.
6. Track state before relying on fluent continuity.
7. Test the artifact instead of trusting the description.
8. Record failure and convert repeated failure into a control.
9. Keep public material separate from private production systems.
10. Require target-specific human approval for consequential external action.
11. Make the strongest claim the evidence supports and stop there.

Before moving into AI-assisted systems and documentation work, I spent more than twenty years in food service and hospitality.

Professional kitchens and service operations taught me that:

• timing is part of quality;
• vague instructions fail under pressure;
• handoffs matter;
• sanitation and safety cannot be implied;
• a beautiful plan is worthless if an operator cannot execute it;
• failures must be recovered from immediately and examined later;
• systems matter more than polished language.

I hold three diplomas from Algonquin College of Applied Arts and Technology in Ottawa:

• Culinary Management;
• Business Management and Entrepreneurship;
• Hotel and Restaurant Operations Management.

That background now feeds the way I build documentation, workflows, review systems, behavioural controls, and practical operating structures.

The strongest fit is remote, document-first, result-based work involving:

• AI workflow governance or responsible-AI operations;
• AI evaluation, claims auditing, or evidence-bounded review;
• human oversight and approval architecture;
• behavioural control and continuity through external operating systems;
• documentation and knowledge operations;
• model evaluation and quality review;
• constraint-first opportunity or workflow evaluation;
• workflow mapping and failure analysis;
• standard operating procedure development;
• source, provenance, and traceability controls;
• public-safe conversion of complex internal systems;
• specialist support for AI governance, privacy, compliance, cybersecurity, accessibility, documentation, or transformation teams;
• unusual project work where conventional job titles do not describe the actual problem.

I am especially useful on a bounded first engagement: one process, one pain point, one current-state map, one set of controls, and a result that can be inspected.

Credibility depends on stating what the work is not.

I do not claim:

• machine-learning engineering experience;
• model-weight modification through external control documents;
• hidden access to model internals;
• production cloud or model deployment expertise;
• legal, audit, clinical, or regulatory authority;
• enterprise implementation where the evidence is internal design and operation;
• that artificial intelligence output is reliable without review;
• that a title can substitute for inspectable proof;
• guaranteed business outcomes before they have been measured.

I can work alongside technical, legal, privacy, security, compliance, preservation, or domain specialists by building the operating and documentation layer around their expertise.

For remote contract work, project collaboration, artificial-intelligence evaluation, workflow governance, behavioural control, documentation architecture, operational writing, or systems work:

info@paranoidpeoplelivelonger.com

Public portfolio: PPLL Signal Archive
Website: Paranoid People Live Longer
Location: Ottawa, Ontario, Canada