Haris MekicAI consultant and operations practitioner
Operations experience.
Applied through AI.
I’m Haris. I work where a plan has to become a result. Programme delivery, budgets and coordination taught me to connect the people, information and decisions behind the work. That is the experience I bring to building with AI.
When experience is scattered across documents, the first challenge is knowing what you can trust and reuse. I built a local evidence system around that problem. I am applying the same thinking to bids, budgets and project decisions.
I am building toward a digital office where people and AI agents share the knowledge, understand their responsibilities and carry work forward together.
MEKreflect is the company I started to pursue that direction. This is my personal portfolio of the work behind it and what I want to develop with a team.
- Operational experience
Understand the people, budgets, deadlines and decisions behind delivery.
- Structured context
Connect source material, responsibilities and the questions that need an answer.
- Working implementations
Build evidence records, bid workflows and tools whose behaviour I can inspect.
- Next, measured pilots
Compare one workflow with the existing approach and measure the full effort with its users.
Selected work
The work behind the vision.
I want you to see the problem I understood, the decisions I made and what came out of the work. Each case has its own evidence and its own next step.
The problem
My experience was spread across documents, correspondence, old CVs and project records. Each new application meant reconstructing the work again. A fluent description could still miss the responsibility or repeat a claim the original evidence did not support.
My contribution
I defined the source hierarchy and the distinction between what I coordinated, what I approved and what still needed confirmation. I directed the Python and SQLite implementation with coding agents, then challenged the documents against the records. The useful change was preserving knowledge between tasks, not adding more chat windows.
What goes in
A defined application or professional question, approved local source material, reusable claim records and the privacy rules for that output.
What comes out
An evidence-linked set of statements, a tailored document package and a persistent record of preparation, review and outcome. The same professional facts can support another document without being reconstructed from conversational memory.
Acceptance checks and the failure to watch
What an accepted result needs
A factual statement names an allowed claim and evidence registered to that claim.
A missing, unrelated or changed receipt cannot establish confirmed completion.
Private source records stay outside the public portfolio and its assistant.
Where it can go wrong
A well-written sentence can still introduce unsupported authority. A fixed-rule validator catches configured patterns. I still review the meaning against the sources. A file hash checks integrity, not the truth of the underlying claim.
The idea behind SOAP
The context is part of the engineering.
When I work with someone, I start with their goals and the work they actually need to deliver. Together we decide what the AI needs to understand, which sources it can use and what a useful result should contain. The person learns to direct the work and challenge the answer. The working method improves through that feedback.
I turn that understanding into contextual standard operating procedures. Each role gets its purpose, relevant sources, permitted actions, expected output and handoff conditions. The next role receives the work, supporting evidence, decisions and open questions, not another conversation to reconstruct.
- ContextObjectives, sources, constraints and authority
- RolesSpecialist responsibilities and bounded tools
- HandoffsStructured outputs, checks and decision records
I enjoy that process of discovery with Claude, Codex and specialist workflows. My experience helps me ask better questions and challenge what comes back. I want a team to learn with the system, improve its procedures and remain responsible for the decisions it makes.
Ask Evidence
Explore the work behind the words.
Ask a specific question about my experience or a published workflow. The assistant retrieves public material and returns an AI-generated answer with references.
Cloudflare Workers AI processes your question and selected public context. Do not include confidential information.
Your question is sent when you press Ask.
How this assistant works
The assistant uses only the curated public portfolio. It cannot access private files or act for me. An AI-generated summary is not independent verification.
This application stores no raw question history. The shared allowance is 20 calls per UTC day. A failed or cancelled model request can still use one reservation.