Workflows with clear boundaries
Turn a research, document, or content task into defined steps, structured outputs, and human review points. Use a simple automation when an agent is not needed.
AI agent engineer / full-stack developer
I help small teams build AI workflows, connect tools through MCP and APIs, and get existing software working. We start with one useful task, clear boundaries, and a way to check the result.
Code, checks, and handover notes are included.
I combine full-stack development with AI workflows that read information, use tools, and produce work someone can check. The job includes the software around the model, not just the prompt.
Turn a research, document, or content task into defined steps, structured outputs, and human review points. Use a simple automation when an agent is not needed.
Connect AI clients to document publishing and existing APIs. Define tool inputs, access permissions, and explicit approval before publishing or changing external data.
Check outputs against source material and sample tasks. Add bounded retries, usage limits, failure reporting, and repeatable tests before expanding the scope.
A useful agent needs a clear task, limited access, and a way to tell when it is wrong. I do not promise unrestricted autonomy.
See current AI engineering workA few projects from my practice. Client details have been anonymized.

Incident recovery
Reproduced the failure in staging, isolated incompatible paths, and built a controlled recovery and release sequence.

Payments and back office
Mapped checkout and callback verification, implemented traceable failure paths, and prepared reconciliation views for operations.

CRM automation
Mapped source fields, added duplicate handling and assignment rules, and created a QA handover for production.
Deliverables, acceptance criteria, and exclusions are agreed first.
Work is delivered through focused code changes, tests, build results, and handover notes.
The goal is to remove the blocker, not turn every task into a rewrite.
Small, focused projects. Scope and price agreed before work begins.
Diagnosis, implementation, tests, and a focused PR for one reproducible issue.
Discuss this projectTypical delivery
Not included: Unlimited investigation or architecture rewrite
Create an automated baseline for a critical journey or remove a build, CI, or release blocker.
Discuss this projectTypical delivery
Not included: Full test-suite coverage
Build one bounded AI workflow or MCP tool integration for an existing product, with defined inputs, outputs, and approval points.
Discuss this projectTypical delivery
Not included: Model fees, model training, unrestricted autonomy, and full multi-agent platforms
Implement localized routes, metadata, canonicals, sitemaps, structured data, and a maintainable content flow.
Discuss this projectTypical delivery
Not included: Large-scale human translation or content operations
Expected behavior, actual behavior, reproduction steps, and any repo or demo you are authorized to share.
Receive written deliverables, acceptance criteria, exclusions, and a quote.
Changes are made in a focused branch or PR and checked against the agreed criteria.
Receive code, verification results, merge or deployment notes, and known follow-ups.
For problems whose root cause is unclear and cannot yet be quoted responsibly.
For one bounded outcome with a clear acceptance path.
For recurring fixes, release support, and gradual technical-debt cleanup.
No unlimited revisions, 24/7 on-call promise, or guaranteed outcome before the system is inspected.
A paid diagnosis can come first. Fixed-price implementation begins only when the reproduction path, relevant modules, and acceptance criteria are clear enough.
You can state the urgency, but acceptance depends on current availability, access conditions, and risk. No deadline is promised before the scope is reviewed.
An NDA can be reviewed when the project requires it. Code, business data, and screenshots are not published without authorization, with or without an NDA.
The payment method, milestones, acceptance window, and any refund conditions are confirmed in writing before work begins. Small bounded tasks may require prepayment or platform escrow.
Each scope includes a written acceptance window. New features and issues outside the agreed scope are quoted separately.
Tools may assist research, testing, and implementation, but delivery is judged by working code, test results, and human review. Client secrets are not submitted to unauthorized third-party services.
Share only information and links you are authorized to disclose. I will first confirm whether the work fits a bounded sprint.