Professional localization work and current technical builds, reflecting both experience and technical direction.
Technical builds and ongoing work focused on models, prompting, and applied AI workflows.
Beta
Built a local-first Document AI system that converts PDFs into structured JSON, indexes document-aware chunks, and answers multi-document questions with page-level citations. It supports native-text extraction, OCR fallback, quality diagnostics, a web UI and CLI, and reproducible retrieval benchmarks.
It turns messy document collections into inspectable, queryable data while validating extraction and retrieval quality instead of trusting unverified outputs.
Ongoing
Built a Chrome extension that analyzes YouTube video transcripts against a user-defined goal using Gemini.
Helps evaluate long-form video before investing time, turning transcripts into goal-based relevance signals.
Ongoing
Building a weather-aware outfit planner for runners, walkers, and everyday outdoor plans. The app combines live forecast data, activity timing, return-home conditions, and personal comfort feedback into a clear clothing recommendation.
Shows applied AI product engineering: deterministic recommendation logic selects the outfit, while AI explains the already locked result instead of inventing advice.
Active
Built a local-first Obsidian operating system with 120+ structured Markdown notes, modular context layers, daily/weekly/monthly review loops, and AI-readable vault organization for planning, journaling, and strategic self-management.
Turns raw personal context into review loops, priorities, and behavioral protocols by combining qualitative reflection with Todoist/Rize productivity signals, recovery inputs, and custom OpenCode reflection skills.
Completed
Built a small Python tool on top of the Polar AccessLink API that exports daily recovery metrics (sleep, HRV, respiration, ANS) into a clean CSV dataset. The data is used in my AI journaling workflow to analyze how training, sleep, and recovery affect day-to-day performance.
Creates a reliable personal data feed for downstream analysis instead of manual recovery tracking.
Completed
Hands-on Python experiments focused on prompting, structured outputs, and practical intuition around model behavior.
Builds working intuition for LLM behavior through small, testable notebooks rather than abstract prompt examples.
Work focused on localization, multilingual UX, terminology, and supporting product systems.
Ongoing
Managing Polish-language localization across Crypto.com's Web3 platform, including website and mobile wallet. Focused on terminology consistency, UX clarity, and trust-building across high-risk financial content.
Sustained quality ownership across high-stakes fintech surfaces where precision affects trust, comprehension, and conversion.
Ongoing
Reviewing and proofreading Polish versions of the OKX website and mobile app, improving linguistic accuracy, terminology consistency, and user trust across exchange and wallet-related flows.
Improved product-language clarity across exchange and wallet content with a focus on risk-sensitive financial terminology.
Ongoing
Applying translator-level precision to multilingual product content, terminology systems, and QA workflows to improve release quality and content consistency.
Connects linguistic review with repeatable QA patterns for terminology, release checks, and product consistency.
Projects focused on analysis, structured data, and evidence-driven decision making.
Completed
End-to-end analysis of Fitbit user behavior data using R and tidyverse, translating activity and wellness patterns into structured business recommendations for the Bellabeat platform.
Converts behavioral data into product recommendations, connecting analysis output with business-facing decisions.