AI Product · Product Design & Implementation

NUS-ISS MTech in Artificial Intelligence Systems · Penn State Computer Science

I build AI product prototypes and work through the choices behind them: what problem matters, what the system should do, and where people need to stay involved. My work spans camera experiences, business agents, learning tools and voice interfaces.

Available full-time for an internship · March–August 2027

LinkedIn · Experience & contribution details

Lightmeta · Product Intern · May–August 2026

Led retro sparkles and hand-tracked star collage effects from concept, interaction rules and PRDs through review and launch. Also proposed and drove iPhone Camera Control. The key choice was to support capture → finished photo → immediate sharing, weighing new effects against latency and computation cost.

LumaQ on the App Store lists sparkles, collage and Camera Control in 1.3.0. My contribution and product choices · FrameText: independent MediaPipe prototype

NUS team project · In development · August 2026–present

I lead business and interaction design, frontend delivery and the backend integration needed for the shop-owner workspace. I defined a first scope of one shop and one product: assess stock, simulate options, compare cash needs, confirm purchases and follow up as conditions change. A simulation can be saved without changing real inventory or cash; purchases need individual confirmation.

The workspace supports simulations, saved results and follow-up tasks. Inventory and cash behavior have been checked with simulated purchasing and arrivals; full autonomous-agent acceptance and outcomes for real merchants remain unverified.

Product walkthrough → · Team repository · Example delivery: workspace results and comparisons

Two-person project · Desktop prototype · September 2026

I lead product design, most implementation and testing, working with AI and a teammate. The agent follows discussion in the background, chooses text or diagrams, and prepares collaboration drafts. The host reviews and distributes a draft; participants respond for themselves. Meeting facts, personal scenarios and confirmed decisions stay separate, with source quotes and versions.

The prototype has engineering and synthetic-flow validation. Continuous updates with real models and the complete real-meeting workflow still need improvement and acceptance.

Screenshots & quick review → · Product choices · Validation & limits

Personal prototype · Original project: January–April 2026

I defined and iterated a voice-first, paper-and-pencil coach for Chinese, maths and English, with substantial AI-assisted implementation. I shaped subject and age-band prompts, used self-testing to find false approval and mechanical questioning, and kept RTC plus System Prompt for response speed and teaching control. Maths leaves the calculation to the learner; Chinese offers structure without writing the answer; English explains a rule for the learner to apply.

The current public walkthrough adds later product and implementation explanations, including RAG and a maths practice slice. August–September additions are separate from the January–April work. The browser demo uses preset content, has no live AI, and does not establish children's learning outcomes.

No-install product walkthrough → · Maths practice · Source & subject scope · Product decisions

Penn State industry-sponsored capstone · Three-person team · August–December 2025

I led alpha/beta bot development and integrated Azure speech recognition, prompting, response handling and speech playback with a teammate's fine-tuned model. Hardware cost, device APIs and the semester timeline led the team to a laptop voice prototype. I contributed to that scope decision, debugging, timing and testing.

Delivered a laptop software prototype; physical glasses and industrial deployment were outside the delivered scope. Sponsor documents and the team report remain private.

Responsibilities, trade-offs & reported results →

• FoodLens SG — Restaurant research with explicit delivery and budget checks before choosing.
• DecisionTrace — Review product commitments against code evidence; semantic suggestions remain advisory.
• Stock Portfolio — Exact daily portfolio calculations before optional AI interpretation; source and synthetic examples.
• Codex Notch — An unofficial macOS companion for noticing actionable coding-task changes.
• Nianxing · 念行 — Capture a thought, review the draft, and confirm what gets saved.
• FrameText — Independent camera-effects exploration with palm tracking and preview-to-export consistency.

Earlier experience: mobile OS issue reproduction at Xiaomi and Python debugging instruction as a Penn State teaching assistant. Details

Mandarin · English | Connect on LinkedIn →