I build from the work people need to do.
My background is in institutional operations, EU project delivery, budgets, procurement and coordinating experts. I use that experience to ask practical questions of AI: what context does it need, which decision can it make, what must be checked, and can the next person continue the work?
I founded MEKreflect to develop those questions through applied research, implementation and collaboration. I define the problem, work with Claude and Codex to build and inspect software, and review the results. The code is AI-assisted. The operational direction, the questions and the responsibility for what I publish are mine.
SOAP means Super Operations AI Procedures. I bring source knowledge, role-based instructions, specialist workers, explicit handoffs and human review into the same process. I want people to become more capable with the system, including knowing when to challenge its answer.
My current research examines context loss, conflicting source versions, decision authority and valid work that gets blocked unnecessarily. The operational evaluations show recorded experiments and their limits. The research observatory connects attributed external model measurements with these more specific operational questions.
| Repository | What is available |
|---|---|
| MEKreflect research labs | OBSERVER8 context-handoff evaluators, operational controls, visual production, telemetry parsing and evidence-expiry gates. Source provenance, 113 Python/Node tests and 17 Java assertions, including actual video encode/decode. |
| MEKreflect research platform | The research website, attributed AI benchmark importer, operational evaluation records and Cloudflare backend. A clean source release with local tests. |
| ParrotCore research | A Python semantic-behavior runtime, simulated creature profiles, trajectory and fault models, with tested installation and package resources. |
| AI enablement workbench | Runnable offline examples of evidence retrieval, receipt checks, context survival and provider-neutral messages, with a source map distinguishing existing code from adapted examples. |
| SERVARI reference shell | A local model interface, human verification queue, bounded action runner and synthetic demonstration data. Includes verification and tests. |
| Architecture studies | Historical design notes and deterministic simulations with explicit assumptions. These are scenarios to inspect, not measured business returns or an independent audit. |
I distinguish an implemented mechanism, a local test, a recorded model experiment and a result someone has achieved in their work. Those are different kinds of evidence. Each project should make clear which one it provides.
The workbench's 16 offline tests were rerun on 16 October 2026. The research platform passes 1,525 release assertions, 56 case-study checks, homepage data-contract checks and 12 SQLite API tests in its public inspection copy. The research labs pass 113 Python/Node tests and 17 Java assertions. ParrotCore passes 56 software tests, including installation into a fresh environment; physical hardware validation is separate. The SERVARI CI checkpoint completed successfully on 16 October 2026 against commit e9a6fbf. Its Python verification and UI checks are an implementation checkpoint, separate from live-model evaluation.
I am interested in research and delivery teams that want to connect AI capability with the reality of organisations, especially the people, procedures and information that determine whether a system is useful.
MEKreflect research and consultations · Professional portfolio · Research reading desk
Company enquiries: info@mekreflect.com
Direct contact: harism@mekreflect.com