Ege Bostancı

Ege Bostancı

AI Operations · Support Operations · Workflow Automation

I build safer AI workflows for support operations.

I combine hands-on B2B support experience with automation and measured AI systems — using strict contracts, deterministic fallbacks and human review instead of black-box autonomy.

Read the AI Support Operations case study → View CV Discuss a role or workflow →

Open to remote roles

One request, step by step

From a support request to a reviewable draft.

Recorded example from the Python rule-based baseline, using synthetic test data.

  1. A request arrives

    Urgent: the synthetic demo service has been unavailable since ten minutes ago.

  2. The system prepares triage

    Category
    Service disruption
    Priority
    High
    Security warnings
    None flagged in this example

  3. A person reviews it

    The output requires human review. The operator checks the classification and draft before deciding what to do.

    No reply is sent and no service action is taken by the copilot.

Read the suggested reply
Draft for human review: We received your service disruption request. A support operator will review the details before any action is taken.

This is the baseline's fixed reply template.

Inspect the synthetic test case →

AI Support Operations Copilot · Synthetic test results from 4 August 2026, not production accuracy. Category scores cover 33 paraphrased requests per model, including rejected calls as failures. Review means a required output flag, not completed human reviews. See all four model results →

Operating principles

Systems that remain reviewable when the model is wrong.

Constrain the AI step

Strict response contracts, validation and fail-closed handling before output can move.

Keep deterministic paths

Rule-based baselines and fallback behavior make failure visible instead of mysterious.

Design for operators

Human review, explicit limitations and documentation are part of the system, not cleanup.

Flagship case study

AI Support Operations Copilot

Prototype completeNot deployed

A human-supervised operations copilot for synthetic support requests: classify the request, detect urgency and missing information, flag security-sensitive content, and keep every result behind review.

  • AI Support Operations Copilot · Synthetic test results from 4 August 2026, not production accuracy. Category scores cover 33 paraphrased requests per model, including rejected calls as failures. Review means a required output flag, not completed human reviews.
  • Nonconforming model output rejected before downstream use
  • No customer data, autonomous decisions or external actions

Read the full case study → Review the repository →

correct categories across four models
23–33 / 33
unvalidated outputs passed in 308 test calls
0
returned outputs required human review
100%
offline automated tests
94

Selected work

Other systems and operational case studies

View all projects →

ShiftScan

Live demo In development

Turns photographed shift rosters into validated calendar files, with optional planning and no persistent server-side storage.

Project details → Live demo →

Quiet VPS Watchdog

Retired

A deterministic readiness check that suppressed repeat alerts, reported recovery once, and never put an LLM in the check path.

Project details →

Writing

Notes from systems I operate

Turkish writing about AI operations, self-hosting, automation and maintenance decisions, plus a linked legacy archive from 2012–2019.

Türkçe sayfa → Yazılar →

Contact

Discuss a role or workflow.

Remote AI operations, technical operations and focused workflow-automation work.

[email protected] → LinkedIn → Services for teams →