The word that best describes me: system builder.
I am a data analyst based in Istanbul. I studied history; a path that began in graphic design moved into media analytics and then into data- and AI-driven reporting systems. Those three layers define how I work: read an event in context, make it visually legible, and tie it to a repeatable system.
I am not satisfied with preparing a good report; I build the process that produces it, measure it, and make it independent of any one person. The approach is source-agnostic: social media, news, customer reviews, internal data — the same discipline applies.
For more than six years I have produced digital media monitoring reports for brands, institutions and public figures. Over the last year at Dedecta (a digital media monitoring, analytics and software company) I rebuilt the entire analysis and reporting process: the path from raw data to delivery is now largely automated, the AI classifier is bound to a measurable test protocol, and the team's knowledge lives in a decision-record system that no longer depends on individuals.
Working principles
- Output must be usable as delivered; a "draft" is not a deliverable.
- The standard never drifts: every period of a report family is indistinguishable in consistency from the client's point of view.
- Rules come from records, not memory: every threshold and exception maps to a decision record.
- AI is not hired until it is measured: precision and recall are reported separately, period by period.
Five skills
Click oneData analysis and executive reporting
Cleaning multi-channel datasets, relevance control, pivot and breakdown generation, executive summary writing. Weekly, monthly and crisis reports for holding-level management, marketing and corporate communications teams.
- Pivots
- Period comparison
- Executive summary
- Crisis report
Social listening and topic taxonomy
Keyword and query design for brands, institutions and individuals; two-level (L1–L2) topic taxonomies; sentiment and tone labelling; filtering same-name people and confusable entities; competitor and sector benchmarking.
- Query sets
- L1–L2 taxonomy
- Same-name filtering
- Competitor benchmarking
AI classification and prompt engineering
Prompt design for LLM-based spam/relevance and sentiment classifiers, layered regression test sets, reason-code architecture, precision/recall measurement and version control. Evaluation frameworks that score third-party AI providers.
- Prompt v2
- Regression set
- Precision / recall
- Vendor scoring
Automation and report-generation pipelines
Python pipelines that take raw CSV → clean data → pivots → PowerPoint; chart and card engines; QA audit scripts; scheduled report delivery via browser automation (WhatsApp / e-mail).
- Python pipeline
- python-pptx
- QA scripts
- Playwright
Knowledge management and system design
Decision-recorded, linted knowledge bases (Obsidian) that make team knowledge independent of individuals; schema, router and lookup layers that AI agents can operate on; product requirement documents and test protocols.
- Decision records
- Router / lookup
- Linter
- Product definition
Experience
- Aug 2025 – presentData AnalystDedecta — digital media monitoring, analytics and software, Istanbul
- 2021 – 2024Analysis & Reporting Specialist (Senior)Papillon Istanbul — media monitoring agency
- 2020 – 2021Analysis & Reporting Specialist (Junior)Papillon Istanbul
- 2016 – 2017Graphic DesignMihmandar Magazine
- EducationIstanbul University, Faculty of Letters, History
- LanguagesTurkish (native), English (upper-intermediate), Arabic (basic)
Toolbox
- Data
- Python (pandas, openpyxl, python-pptx), Excel (advanced / VBA), SQL, Power BI
- AI
- Claude, ChatGPT, prompt design, regression test sets, LLM evaluation
- Automation
- Playwright, Node.js, macOS launchd, shell
- Knowledge
- Obsidian, YAML decision records, Markdown
- Monitoring
- Enterprise social listening platforms (hands-on with 3)
- Presentation
- PowerPoint template engine, chart XML editing