Tshivhidzo Mbedzi AI & Technology Specialist

Hire me

What you can actually buy, and how to start.

Six things I can be engaged to do, each one backed by work on this site rather than by a claim. Available for a paid trial task, contract work, part-time hours or a full-time remote role.

Open to remote roles worldwide · Available immediately · SAST (UTC+2)

Ways to work together

Four ways in, smallest first

Start here

A paid trial task

The lowest-risk way to find out whether I am any good: a small, scoped, paid piece of real work. Rate a set of outputs, review a document, test a feature, write up a process. You see the actual quality instead of reading about it, and you are out very little if the answer is no.

Best if you are unsure, or if my lack of formal employment history is the hesitation.

Contract or freelance, by project

A defined piece of work with an agreed scope and deliverable — an evaluation set built and scored, a workflow automated, a body of content reviewed.

Best for a specific problem with a clear finish line.

Ongoing part-time

A fixed number of hours a week on continuing work: rating queues, QA passes, support cover, documentation upkeep.

Best for work that recurs rather than ends.

Full-time, remote

What I am primarily looking for. South Africa, UTC+2, which overlaps the full European working day and the first half of US Eastern.

Best if the work above would be continuous anyway.

Services

Six things I can be engaged to do

Each lists what you actually receive, and links to the work that shows I have done it before. Nothing is offered here that I cannot point at prior evidence for.

AI output evaluation and quality rating

Rating model outputs against your guidelines: accuracy, helpfulness, safety, tone, instruction-following. Written rationales for each judgement, not just scores, so your team can see the reasoning and disagree with it.

What you get

  • Scored outputs with a written rationale per item
  • A list of the disagreements and edge cases your guidelines do not yet cover
  • Notes on where the guideline itself is ambiguous

Evidence

Six documented cases of catching errors in my own results

Content review and fact-checking

Checking written or generated material for accuracy — and for the harder failure, material that is accurate and still leaves the reader with the wrong conclusion. Every figure traced back to its source.

What you get

  • Line-level corrections with the source for each
  • Flagged claims that are true but misleading as presented
  • A note on anything I could not verify, rather than a silent pass

Evidence

Caught a chart in my own material that was accurate and taught the opposite of its lesson

Evaluation design and test harnesses

Building the thing that decides whether your output is good enough to ship: a rubric, a held-out set, an automated gate in your pipeline that fails the build rather than warning about it.

What you get

  • A written rubric your team can be calibrated against
  • An automated check that exits non-zero on failure
  • Documentation of what it does and does not catch

Evidence

A twenty-three-rule gate, including accessibility, that blocks this site from publishing if anything is broken

AI workflow automation

Taking a repetitive process and making it run unattended, with a check at the end that can refuse to deliver. Generation, verification, assembly and publication.

What you get

  • A working pipeline you can run yourself
  • A verification step before anything is released
  • A runbook for when it fails at three in the morning

Evidence

Twenty 4K episodes produced unattended, behind an automated delivery gate

Technical documentation and runbooks

Writing the documentation that gets read: architecture notes, deployment runbooks, honest limitation statements, onboarding material. Written for the person who has forgotten, which is usually the author.

What you get

  • Documentation that someone unfamiliar can actually follow
  • An explicit statement of what the system does not do
  • Diagrams where a diagram is clearer than a paragraph

Evidence

This entire site, its README and its deployment guide

IT support and troubleshooting

First-line and second-line support, systematic fault diagnosis, and the operational work around running systems that have to keep working. Backed by a formal qualification rather than enthusiasm.

What you get

  • Faults diagnosed to a root cause, not worked around
  • Written resolution notes so the same fault is cheaper next time
  • Honest escalation when something is beyond me

Evidence

Higher Certificate in IT Support Services, and a GPU workstation kept running under sustained load

Terms

Rates, and what I will not pretend

Rates depend on the work, so I would rather quote against a real scope than publish a number that is wrong for both of us. Tell me what you need and I will give you a figure and an estimate.

Three things worth saying before you ask

  • I have no commercial client references yet. You would be the first, and the paid trial task above exists precisely because of that.
  • I build software with AI assistance and say so. If that is a problem for your organisation, better that we both know now.
  • If something is outside what I can do, I will tell you rather than take the work and find out.

Send me the scope and I will tell you honestly whether I can do it

Including when the answer is no. If a paid trial task is the easiest way to find out, propose one — that is what it is there for.