Portrait of Chung-Chun Wang

Full-Stack · Cloud Infrastructure · Applied AI

Chung-Chun Wang

Chung-Chun Wang · Takala Wang

Software Engineer / Full-Stack & Cloud Infrastructure

Résumé PDFEmail ↗LinkedIn ↗GitHub ↗Blog ↗

About

I am Chung-Chun Wang, a full-stack and cloud infrastructure engineer and open-source maintainer who turns research prototypes and AI applications into reliable production systems.

  • National Yang Ming Chiao Tung University

    Master of Science, Institute of Computer Science and Engineering

    2025–2027 · Currently enrolled

  • National Taiwan Normal University

    Bachelor of Science in Computer Science and Information Engineering

    2021-09–2025-06

01

Experience

  1. 2025-12–Present · Part-time

    GitRoll — Full-Stack Engineer / R&D

    Owns the AI-collaboration scoring in the AI Skill Engine and the end-to-end Skill Challenge flow, from question design and the answering environment to scoring and reports.

    • AI Skill Engine: an LLM grader you can verifyOn real data, the gap between model and human scores falls within the gap between two human raters; the design trade-offs and validation are recorded as ADRs.
    • Skill Challenge: end to end, from question design to reportsDelivered the full flow from question design and answering environment to scoring and reports, with automatic submission intake and sync through Google Classroom.
  2. 2025-07–2025-08

    Acer Medical — R&D Summer Intern

    Took the aiGait gait-analysis model from the development environment into real mobile/edge use, making inference about 300% faster.

    • The model is a cascade that detects objects first and then estimates the skeleton; I moved inference from PyTorch to ONNX Runtime.
    • Moved pre- and post-processing onto the device and vectorized it to cut overhead.
  3. 2023-07–Present

    Cool English — Part-Time Full-Stack Developer

    Long-term development of three English-learning products with the NTNU Department of English and Taiwan's Ministry of Education; I led the Speech Examination Platform from scratch (see Major projects).

    • Voice Studio: Teachers configure multiple characters, voices, pauses, tone, speed, and volume in a visual SSML editor to produce listening-test material directly. I built parts of the frontend, API integration, and audio processing, connecting SvelteKit, Tailwind CSS, and Azure TTS output; it cut manual recording work by about 80%.
    • English Scenario Chatroom: Spoken-English practice in everyday scenarios such as food, clothing, housing, transport, education, and leisure, connecting voice input, chatbot dialogue, and speech playback. I integrated Azure Speech Recognition and Azure TTS on top of Hugging Face Chat-UI; it has been used in 10+ interactive practice sessions.

02

Selected projects

NOJV problem library after sign-in

Open-source online judge

NOJV

An open-source online judge I founded and maintain. I designed everything from problems, courses, and contests to the sandbox, judge scheduling, and Kubernetes releases, and 300+ students use it in live courses, with up to 100 online at once and 2,000+ submissions so far.

  1. 01

    Judge scheduling: replacing a custom coordinator with Temporal's native priorities

    In a load test, 100 submissions were all AC within 243 seconds; after deleting the pods of Temporal's 4 component types one by one, 32 submissions were still all AC with no lost work.

  2. 02

    Sandbox: measuring time and memory accurately

    An empty program now measures 0.98 ms, and the memory-limit smoke test went from about 1/3 failing to 6/6 passing. I later restricted IPC with seccomp and cleared scratch space before and after each test case, closing state leaks between test cases within a stage.

  3. 03

    Release and reliability

SvelteKit · TypeScript · Tailwind CSS · Monaco Editor · Temporal · PostgreSQL · Prisma · Redis

Read the case study →
Hinagiku home page

Collaborative learning platform · 2024-09–2025-06

Hinagiku

A research project with Academia Sinica's Institute of Information Science and National Taiwan Normal University that connects teacher templates, student discussion, live transcription, AI guidance, and outcome analysis through Think-Pair-Share, and has been piloted in real classrooms.

  1. 01

    Owned the LLM chat core, connecting discussion content, prompts, and model responses.

  2. 02

    Implemented discussion/summary APIs that connect discussion data, AI analysis, and summary generation.

  3. 03

    Completed PDF parsing, voice records, and PDF/DOCX export so teachers can produce preservable outcome documents directly from classroom discussion.

SvelteKit · TypeScript · Firebase Auth · Firestore · Google Cloud Storage · Cloudflare R2 · Genkit · Google Gemini

Read the case study →
Speech Examination Platform production home page

Cool English project

Speech Examination Platform

An English speaking mock-exam platform I led from scratch, covering seven task types, student recording, AI scoring, and administration, which has supported 1,000+ students preparing for exams.

  1. 01

    Async scoring: surviving rate limits, restarts, and simultaneous exam starts

    A load test at the scale of 2000 sessions had 0 failures. The same round caught a problem where, with 100 students starting an exam at once, about 50% of requests failed; it was fixed by switching to READ COMMITTED with a unique constraint and retries.

  2. 02

    Recording reliability: finding the real cause with audit data

    Audit data showed 165/291 WebKit webm recordings were only 5 bytes, disproving the original "cold start" hypothesis; switching WebKit to mp4 fixed it and identified 47 affected iPad answers from the exam.

  3. 03

    Scoring fairness: keeping speech-recognition errors from becoming violations

    Speech-recognition errors no longer flag an attempt or lose its score, and scoring-rule changes can be applied directly to existing results.

SvelteKit · TypeScript · Azure Speech · Azure OpenAI · PostgreSQL · Prisma · BullMQ · Redis

Read the case study →
NYCozU course search (coz-planner)

2026-04–Present · Student open-source community (not formal employment)

NYCU LIFE — Campus platforms and Kubernetes infrastructure

Helps build campus services and a Kubernetes cluster in a student open-source community, owns operations work, and founded a course-search site.

  1. 01

    Zero-trust secret bootstrap for production

    No temporary privileged account remains in production, and the full secret lifecycle is documented.

  2. 02

    Production accepts only versioned images

    Production images must use a SemVer version (optionally with a digest), and non-compliant manifests cannot be merged.

  3. 03

    Services I lead

Kubernetes · Argo CD · Kustomize · Cilium · Traefik · Gateway API · OpenBao · External Secrets

Read the case study →

Research project

LTTC GEPT Speaking Assessment Research

With the Language Training and Testing Center (LTTC) and the NTNU Speech and Machine Intelligence Laboratory, developed an automatic GEPT speaking assessment system that combines pictures, questions, and speech.

  1. 01

    Fine-tuned BLIP-2 and designed prompts for the T5 language model to improve multimodal speaking assessment; accuracy was 71% on familiar content and 68% on unseen content.

  2. 02

    Published at O-COCOSDA 2024 and extended into the data augmentation method of the first-author SLaTE 2025 paper.

  3. 03

    Collaborated with LTTC and the NTNU Speech and Machine Intelligence Laboratory; Advisor: Prof. Berlin Chen.

PyTorch · BLIP-2 · T5 · Whisper · LLM Prompting · Multimodal Assessment · Hugging Face

Read the case study →

Enterprise knowledge retrieval

Delta-NTNU ChatBot

Compares multiple RAG and knowledge-graph retrieval architectures for internal enterprise knowledge queries.

  1. 01

    Designed and compared GraphRAG, LightRAG, Semantic Indexing, and different semantic query methods.

  2. 02

    Built retrieval experiments and an evaluation workflow, analyzed how architecture choices affect retrieval efficiency and answer quality, and organized the results into an enterprise knowledge question-answering system proposal.

  3. 03

    Collaborated with Delta Electronics and the NTNU Speech and Machine Intelligence Laboratory; Advisor: Prof. Berlin Chen.

RAG · GraphRAG · LightRAG · Semantic Indexing · Vector Retrieval · Knowledge Graphs · LLM Evaluation · Retrieval Experiments

Read the case study →

03

Side projects

OnStage TW home page

OnStage TW

A fully static site that aggregates Taiwanese theatre performances with filtering, maps, favorites, RSS, and PWA support.

SvelteKit · Svelte 5 · TypeScript · Tailwind CSS

TWLinter CLI check example

TWLinter

A Taiwan Traditional Chinese wording checker and conversion tool.

Rust · Cargo · CLI · Chrome Extension

urtube home page

urtube

Hackathon overall winner: imports YouTube viewing history with user consent and uses AI to analyze interests, helping people find common topics.

TypeScript · LLM Classification · Semantic Embeddings · Clustering

Weave-In home page

Weave-In

A no-signup browser meeting room for up to 8 people, where an AI facilitator detects groupthink and steps in only after participants consent.

TypeScript · React · Vite · Cloudflare Workers

lookline user journey view

lookline

A fashion social-commerce prototype: buying clothes creates a Look that others can remix, pair, and share, and those interactions feed back into recommendation and manufacturing.

TypeScript · Cloudflare D1 / R2 · Drizzle · Vision Model Enrichment

1111 Job Scout local frontend

1111 Job Scout

A search platform over a fixed dataset of 1,218,635 job postings, using BM25 as production retrieval alongside hybrid-retrieval research controlled by verified gates.

SvelteKit · TypeScript · FastAPI · Python

An edition produced by codex-chronicle

Codex Reporter / codex-chronicle

A personalized newsletter plugin that Codex researches, edits, and publishes automatically, adapting content to reader feedback.

Codex Plugin · Static Publishing · Scheduled Workflows · Feedback Loop

Routa local sample itinerary

Routa

A natural-language itinerary system that replans around flooding, road closures, MRT disruptions, or unavailable YouBike bikes.

Next.js · TypeScript · SQLite · Gemini

repo-lens local home page

repo-lens

A full-stack platform that uses Gemini to analyze Git repository structure, technical composition, and development history.

Next.js · Firebase Auth · Firestore · Gemini

behavioral-kit generating a Traditional Chinese question bank

behavioral-kit

A bilingual behavioral-interview Agent Skill that builds question banks for a role, records answers in STAR structure, and produces an assessment report with a hiring recommendation.

Agent Skill · Python · uv · STAR Interview Method

04

Research & recognition

Papers

Awards

Let's talk.

Work, collaborations, or just talking tech — write any time.

ccwangtakala@gmail.com

LinkedIn ↗GitHub ↗Blog ↗

Updated 2026-09-30