Wenxi (Vincent) Jiao

An app for daily tasks, habits and a mood diary.

• Built reminders, goals, flexible habit check-ins and a mood diary, with notifications and ten interface languages. Users can create tasks through Yumi chat.
• Implementation: tested RFC 5545 recurrence rules, including intervals, limits and end dates.

Save new words as study cards and practise them in short reviews.

• Built AI-assisted vocabulary cards in ten learning languages, with meanings, pronunciation and examples. Reviews adapt to what each learner remembers; cards can also be created through Yumi.
• Implementation: migrated from SwiftUI to Flutter, with regression tests to preserve review scheduling.

A walking game that turns your own route into a trail to follow.

• Released an augmented-reality walking game where players record a route and follow a glowing trail through the camera view, collecting virtual objects with sound and vibration cues.
• Implementation: ARKit/RealityKit with motion smoothing, position-jump rejection and tracking recovery.

Share drawings and photo notes with one person on their Home Screen.

• Released a private, two-person space for drawings and photo notes; drafts stay private until sent.
• Implementation: PencilKit drawing, anonymous pairing and WidgetKit, with picture delivery independent of widget refresh.
• 1.1.0 submitted for App Review: Built AI photo editing with local version history and opt-in nearby pairing for compatible iPhones.

Help an event organiser seat people with new conversation partners.

• Used weekly for around 30 participants at an Auckland Japanese-English meetup. Built phone check-in, live table assignments and round controls for the organiser.
• Design choice: a weighted seating algorithm balances languages and table sizes, prioritising new conversation partners between rounds.

Time karate matches and display scores for a community event.

• Built and operated the tool at an Auckland karate community’s first exhibition on 10 October 2026. The instructor requested a copy for future events.
• Implementation: built macOS and Windows apps with separate operator and audience windows. Used monotonic timing and restored saved matches in a paused state; also built an offline browser timer.

Track job applications, interviews and the documents sent for each role.

• Built an organiser for applications, interview dates, notes and résumé versions, keeping the next step and application history together.
• Design choice: save locally so editing works offline; merge independent edits and keep both versions of conflicting fields. Tested offline edits, restarts and conflicts; added versioned attachments.

Party games for a group in the same room, using one phone or several.

• Building Word Mystery and Antidote at Dawn, with private roles and guided turns. Word Mystery asks players to describe secret words and vote for the person with a different one.
• Implementation: offline shared-phone play and connected rooms with private views and shared round state. Seven interface languages and word translations; players use a common spoken language.

Find research papers relevant to a claim and show passages for review.

• Built a paper-search and reading workbench to help researchers find sources behind a claim and inspect the relevant passages.
• Evaluation: compared nine pipelines on 28 curated machine-learning claims. The best retrieved a labelled relevant paper in the first five results for 27/28 claims, versus 9/28 with keyword search (BM25). This small benchmark measures paper retrieval, not whether a passage proves a claim.
• Implementation: combined scholarly search, citation expansion, semantic embeddings and passage-level checks, with PDF/HTML parsing and deduplication. Ranking by citation frequency alone achieved 89.3% Hit@5 before model-based relevance assessment.
• Integration: delivered a command-line tool and eight MCP tools for use by other AI systems. Integrated with my supervisor’s system without modifying its source code.

Train a simulated ship to follow a route and avoid collisions.

• Trained a navigation policy with rewards based on maritime collision-avoidance rules (COLREGs). Achieved 95% collision avoidance after approximately 4,000 training episodes in simulation.
• Simulation: built a four-degree-of-freedom MMG ship model that accounts for inertia and delayed response to controls.
• Implementation: implemented Soft Actor-Critic from the research paper in PyTorch and compared two reward designs.

Volunteer work in community sport and student support.

• Train weekly with a non-profit karate community in Auckland offering free participation. Volunteered as a calligrapher for its first exhibition, lettering participants’ names on certificates.
• Supported new students’ transition to university life as a UniGuide volunteer at the University of Auckland.

Connect with people across cultures.

• Attend Korean, Japanese and English meetups every week in Auckland. I keep learning languages to make friends with people from different cultural backgrounds.

Make guitar practice more immersive with sound-responsive lighting.

• Built a home automation setup with AI voice control for lights and appliances, and programmed lighting to respond to my voice and guitar. I also enjoy busking with guitar and vocals.

Keep learning new skills and stay active.

• In Auckland, I have explored barista training, motorcycle riding, forklift operation and light-aircraft flying. Other interests include running and half-marathons.