Three environments taught me three different meanings of “working.”

Backend work on reviewable AI-assisted clinical workflows, with explicit human-control and audit boundaries.

Work record

• Built a clinician-reviewable referral-draft backend with structured generation metadata, PostgreSQL persistence, and doctor-only edit, approve, and reject APIs.
• Added approval records, immutable audit events, duplicate-review protection, deterministic rule-engine fallback behavior, and 12 focused tests.
• Kept the completed scope explicit: provider-backed generation, referral submission, and the frontend review UI were not part of this draft contribution.
• Continue contributing to the project and conducting a literature review for a MedQX research manuscript.

A year of enterprise workflow testing taught me to separate observed failures from plausible explanations before changing automation.

Work record

• Updated and debugged end-to-end automation for Vault RIM Core workflows, reducing flaky regression behavior.
• Wrote automation scripts, test plans, and test cases using Java, SQL, HTTP, and browser-automation logic.
• Triaged automation and UT/Pre-UT failures across application behavior, rendered text, data state, and cleanup steps, then communicated confirmed evidence and open hypotheses clearly.

Research on adaptive therapeutic media, followed by manuscript work on offline RL for adaptive music visualization.

Work record

• Researched AI-driven art therapy and adaptive media informed by music, visual, and physiological features.
• Supported refinement of machine-learning pipelines for adaptive therapeutic content generation.
• During the final roughly five to six months, drafted and revised manuscript sections and connected literature on art therapy, cross-modal mapping, affective computing, multimodal sensing, and offline RL.