Applied AI Engineer · AI Evaluation · LLM Reliability

I build source-grounded AI systems with agents, evaluation harnesses, safety gates,
observability, and production-quality engineering.

$ pytest tetiana/ -v --tb=short

tetiana/applied_ai.py::test_source_grounded_agents .................... PASSED
tetiana/llm_eval.py::test_output_grounded_in_cited_source ............. PASSED
tetiana/llm_eval.py::test_model_admits_when_evidence_is_missing ....... PASSED
tetiana/platform.py::test_prompt_changes_require_evaluation ........... PASSED
tetiana/experience.py::test_years_in_quality_engineering .............. PASSED
tetiana/education.py::test_bu_masters_ds_ai_ml ........................ PASSED
tetiana/llm_eval.py::test_same_prompt_same_answer ..................... FAILED

========================= 6 passed, 1 failed ==========================

FAILED test_same_prompt_same_answer
  AssertionError: expected deterministic output, got a language model
  → this is why evaluation, observability, and safety gates matter

I build reliable Applied AI systems around the model: source-grounded agents, evaluation harnesses, prompt governance, safety controls, telemetry, and human-review workflows.

My foundation is more than a decade in software quality engineering and automation for data-intensive and financial products. That experience shapes how I approach AI: important behavior should be testable, observable, explainable, and reversible.

Traditional test automation assumes that the same input produces the same output. Generative AI removes that assumption, so I evaluate what cannot be handled by an exact assertion:

• Is the answer grounded in its cited evidence?
• Does the model abstain when evidence is missing?
• Did a prompt or model change introduce a regression?
• Can a failure be traced to its data, prompt, model, cost, and latency?

The goal is not only “does it return 200?” It is “is it right, is it grounded, and can I evaluate it systematically?”

An independent publishing-intelligence platform with research agents, source-backed ingestion, review queues, prompt governance, execution telemetry, AI evaluation, deterministic rights-safety controls, and controlled promotion workflows.

Live platform · Case study · Video walkthroughs

A public evaluation framework that scores generated claims against cited sources. It demonstrates golden datasets, deterministic grading, optional LLM-as-judge evaluation, hallucination detection, groundedness checks, and regression testing.

View code · Evaluation case study

A data-science project comparing regression and tree-based models using cross-validation, RMSE, MAE, R², residual analysis, and model-risk communication.

View code · Case study

Applied AI and evaluation: LLM evaluation, groundedness, hallucination detection, RAG, AI agents, prompt versioning, golden datasets, model-as-judge, safety gates, tracing, token/cost/latency telemetry, and human review.

Data and engineering: Python, SQL, PostgreSQL, REST APIs, FastAPI, Docker, Git, CI/CD, Jenkins, data validation, analytics, debugging, monitoring, and dashboards.

Quality engineering: Java, PyTest, Selenium, Appium, REST Assured, Postman, TestNG, JUnit, Cucumber, Sauce Labs, Zephyr, and TestRail.

Full-time U.S.-remote opportunities in:

• Applied AI Engineering
• AI Evaluation / LLM Evaluation
• AI Platform and Reliability
• AI Observability and Safety
• Product Data Science and advanced analytics

I bring senior software-quality depth to AI systems without treating AI quality as traditional QA. I am especially interested in teams building agents, RAG workflows, evaluation platforms, data-intensive AI products, and trustworthy production systems.

M.S. Data Science, AI & Machine Learning — Boston University, 2026
Massachusetts, USA · U.S. citizen · Open to full-time U.S.-remote roles