ChiaOS activation sequence

Clean setup handed off to the activated system.

Live age build
01 / original setup confirmed
02 / system director online
03 / proof modules indexed
04 / local assistant armed

System status

Ready for inspection.

Route/system
ModeActivated
SignalAI-first
Activated ChiaOS System

Activated system, running live.

Code. Learn. Build. Evolve. ChiaOS is the full-send layer: 3D direction, local assistant logic, mission proof, Realfun operations, and career identity arranged as one product experience.

Primary signal

AI-first

Operating proof

Realfun

Build style

Vibe coding

System Director

Camera operator online.

The fixed 3D layer reacts to the section in view, mission focus, and skill focus. It behaves like a visual director rather than a decorative background.

Active layer

Scroll-guided 3D

Focus loop

Mission + skill cues

Motion

Reduced-safe

Identity

AI-focused Computer Science student building through cybersecurity, practical systems, and AI-assisted development.

Core signal

Creative development, vibe coding, and practical career direction.

system snapshot

Identity, education, and proof density.

Based in Kuching, Sarawak, Malaysia, ChiaOS compresses Chia's current status into scan-friendly modules before opening the deeper mission archive.

Academic core

AI + Cybersecurity

Year 3, Semester 1 / fifth semester out of six (Semester 2, 2026)

Mission files

12 proof tracks

Projects across AI, design, cloud, database, and networks.

Operations scale

Around 130 students supported

Hotline, scheduling, events, and urgent changes.

Operating style

Build with proof

AI-assisted iteration grounded in real evidence.

Swinburne snapshot

CGPA 3.51

Cumulative, as of the last completed semester.

Foundation

Curtin CGPA 4.00

Completed July 2024

Education route

Bachelor of Computer Science at Swinburne University of Technology Sarawak

Double majoring in Artificial Intelligence and Cybersecurity. Expected graduation: Around June or July 2027, if everything goes well. Cumulative CGPA 3.51 as of the last completed semester.

operations module

Realfun turned pressure into operating rhythm.

Part-time Admin / Hotline Handler at Realfun Learning Centre, Miri Branch. May 2023 to Present. Major real-world operations experience coordinating parents, students, teachers, class changes, online support, and urgent communication under pressure.

Case response

Tynker outage response

When Tynker access broke down, Chia handled incoming messages and calls, guided teachers toward backup special projects, and helped classes keep moving.

200+ contacts handled

Around 130 students supported

30+ weekly class slots coordinated

Almost 24/7 support when needed

Communication

Parent, student, and teacher coordination through daily operational channels.

Scheduling

Class replacements, reminders, and updates kept moving without drama.

Signal routing

Turning scattered requests into clear next actions for the right people.

Operating surface

Handled hotline communication with parents, students, and teachers through WhatsApp Business.
Coordinated class reminders, replacement classes, rescheduling, Zoom support, and online class updates.
Tracked attendance, payment/admin details, and progress report support work.
Created Canva posters and announcements for operational updates.
Handled complaints, urgent changes, and aggressive parent communication professionally.
BugCrusherGirlsCodeScience on WheelsSarawak Space Challenge

mission files

Project proof, ranked for signal.

Each featured project has its own artifact. Pick one to inspect it in 3D, then open its case study for the measured results and stated limits.

Full mission archive

Space · NASA Space Apps Challenge 2026

Earth FM

Live prototypeTeam leader, team BoboMimi

A multisensory Earth Information Jukebox prototype that turns a NASA visual and NASA GISTEMP observations into browser-generated sound, with the mapping kept visible.

Earth FM home page with the Climate Spiral record, visual-sound scanner and four Earth stories.
Live site, captured 13 November 2026
Five-page Studio: Home, Listen, Create, Explore Data and About
Live pixel-scan sonification of NASA SVS 5190, the Climate Spiral
Twelve selected January GISTEMP v4 anomalies on a fixed pitch scale
ReactTypeScriptViteWeb AudioThree.jsReact Three FiberPlaywright
Read case study

Problem

Earth visualizations can be engaging but hard for a first-time learner to interpret. The challenge asked for visual frames paired with dynamic, real-time sonification.

Contribution and role

Chia led team BoboMimi, a two-person team with Jacy Chiong Yet Suu, and contributed to development and integration. The delivered app is shared team work, so features here are credited to the team.

Architecture

  • A 48-strip left-to-right scan maps image brightness to pitch, RGB to timbre and position to stereo pan.
  • Bundled GISTEMP snapshots play on a fixed ten-band scale, with dates, values and frequencies shown together.
  • Fonts, visuals and observations ship with the app, so there are no runtime data requests or uploads.

Technical challenges

  • Keeping image sonification clearly separate from measurement: the scan is a visual pattern, not a thermometer.
  • Equal note length does not encode elapsed time, so exact values and gaps stay visible beside the melody.

Results

  • Public prototype with two curated Earth Stations: Climate Spiral and Arctic September.
  • English and Simplified Chinese, keyboard controls and reduced-motion defaults.

Limits: No participant study has been run. Learning effectiveness and screen-reader and physical iPhone behaviour are not yet established.

Evidence: Repository README and docs/SUBMISSION.md, checked 13 November 2026.

Fintech · COS40005 Project A, client: Sarawak Artificial Intelligence Centre

FinGence: Finance Intelligence

In progressData Ingestion Lead and Integration Contributor

AI-driven financial analysis and automated reporting for the Sarawak Artificial Intelligence Centre (SAIC): ingestion, reconciliation, anomaly review and management reports with traceable evidence.

Deterministic CSV/XLSX ingestion with column mapping, validation and lineage
Bank-to-ledger reconciliation with row-level source evidence
Signed-in workspace with roles, saved runs and an audited AI-call gateway
Next.jsTypeScriptSupabasePostgreSQLpdf-libTesseract.jsVitest
Read case study

Problem

Finance teams reconcile bank and ledger records, investigate unusual movements and write management reports by hand. Each step needs evidence a reviewer can trace back to source rows.

Contribution and role

Led the data-ingestion workstream: deterministic CSV/XLSX ingestion, validation, normalization and source lineage. Also contributed to AI-assisted ingestion workflows and application integration. Teammates own the analytics, backend and security components, and this case study describes the team's shared system.

Architecture

  • Ingestion profiles source values, validates rows and keeps lineage and quarantined records for review.
  • Every AI model call passes through one gateway that records metadata only; a test fails the build on a direct provider call.
  • Supabase functions run as the service role only, with forced row-level security and scrypt-hashed passwords.

Technical challenges

  • Keeping original exceptions and financial facts separate from human review decisions.
  • Reporting measured results honestly, including counts that differed from the original brief.

Results

  • Synthetic March PoC: 175 bank and 177 ledger rows sorted into seven reconciliation outcomes.
  • Rule-based anomaly baseline on constructed synthetic data: 86.7% recall, 41.1% precision.

Limits: Results come from synthetic data, not client files, and do not prove accounting correctness. Several AI features stay disabled in hosted production.

Evidence: Repository README, docs/handoff/CURRENT.md and the combined PoC runbook, checked 13 November 2026.

AI · COS30082 Applied Machine Learning, Assignment 1 (individual)

200-Species Bird Classifier

BuiltIndividual project

A fine-tuned EfficientNetB3 image classifier for 200 bird species, packaged as an educational Streamlit app with Top-5 predictions and documented limits.

76.41% Top-1 and 93.69% Top-5 on the official 1,204-image test set
CPU float32 inference at 380 × 380 with verified class mapping
Upload limits, provenance lock and attribution for upstream components
PythonTensorFlow / KerasEfficientNetB3StreamlitTransfer learning
Read case study

Problem

Classify photographs into 200 closely related bird species, then ship the model so anyone can test it without a GPU.

Contribution and role

Individual assignment (AS01): Chia fine-tuned EfficientNetB3 from ImageNet weights, evaluated it on the official test partition and packaged it as a Streamlit app. This is separate from the COS30082 group project.

Architecture

  • One frozen EfficientNetB3 model with 380 × 380 RGB input and CPU float32 inference.
  • The ordered class mapping and preprocessing are verified before the model loads.
  • Uploads are limited to 10 MiB and 20 megapixels, and the app ranks the Top-5 species.

Technical challenges

  • Fine-grained classes: many species differ only in small markings.
  • Communicating that softmax confidence is not a calibrated probability.

Results

  • Official test (1,204 images): 76.41% Top-1, 75.85% average class accuracy, 93.69% Top-5.
  • Internal validation (959 images): 80.29% Top-1, 93.43% Top-5.

Limits: The model only knows the 200 supplied species and will still label other birds or non-bird images. The hosted Streamlit app currently requires a Streamlit sign-in, so only the source is linked here.

Evidence: Repository README, verified results table, checked 13 November 2026.

Product · Supplementary prototype, based on the SWE30003 Assignment 01 case

Sarawak Tourism Information System

Live prototypeUI/UX prototype development

A responsive React UI/UX prototype derived from the SWE30003 draft SRS, for discovering Sarawak, planning a multi-day trip and demonstrating a local listing-management workflow.

Sarawak tourism prototype home page with a rainforest hero image, search bar and experience filters.
Live site, captured 13 November 2026
Search, filter and compare up to three attractions
Day-by-day trip planner with text export and explicit shared-link import
Local management demo with archive, restore, activity log and CSV export
ReactTypeScriptViteResponsive designLocal storage
Read case study

Problem

Turn an incomplete draft SRS for a Sarawak tourism system into a usable visitor and staff experience.

Contribution and role

Supplementary UI/UX prototype work, maintained in Chia's public repository. It maps the draft SRS to working flows and documents design assumptions where the SRS was silent. It is not the completed SWE30003 assessment.

Architecture

  • A browser-only React app; saved places, trips and edits live in local storage.
  • Shared trip URLs carry only catalogue entries and require the visitor to click Import trip.
  • Provider links must use HTTPS and open in a separate tab.

Technical challenges

  • Filling gaps in the SRS without presenting assumptions as requirements.
  • Fitting eight sections into a mobile bottom navigation.

Results

  • Public prototype verified at 1440 × 1080 and 390 × 844.
  • The management session ends after 15 minutes of inactivity.

Limits: A UI/UX demonstrator only: no accounts, payments, real reservations, live availability or shared database. The management area is a local demo and does not meet the SRS's staff authorization requirement.

Evidence: Repository README, checked 13 November 2026.

Product · Personal product, June 2026 to present

PagePilot

Live productProduct development: scope, UX and technical direction

A free, local-first PDF workspace for merging, splitting, reordering, compressing, annotating and signing documents in the browser, with no accounts and no uploads.

PagePilot home page with a local PDF workspace and the Organize, Send and Sign, and Review tool groups.
Live site, captured 13 November 2026
Previews render with PDF.js and exports are written with pdf-lib, in the browser
Public privacy page: no PDF content, file names, sizes or page counts in analytics
Known limits are published: structural compression, flattened notes, visual signatures
Next.jsReactTypeScriptPDF.jspdf-libWeb WorkersVercel Analytics
Read case study

Problem

Students often need to merge, split, trim, sign or shrink a PDF right before a deadline. Many free tools upload the file, add a watermark or require an account.

Contribution and role

Chia's own product: he defined the scope, the user experience, the technical direction and the browser-based PDF workflows. The source is private, so this case study uses only facts from the live product and its public pages.

Architecture

  • Files are read into browser memory. PDF.js renders previews and thumbnails, and pdf-lib writes the exported PDF.
  • Heavier visual reorder, annotation and signature exports run in a Web Worker so the page stays responsive.
  • The result downloads from a temporary local Blob URL. Documents are never stored in cookies, persistent browser storage, an offline database or a service-worker cache.
  • Cookieless analytics count only event names and tool names. Automatic pageviews are discarded.

Technical challenges

  • Measuring whether people complete a task without collecting anything about their documents.
  • Being honest about what browser-only processing can and cannot do.

Results

  • Public beta with merge, split, reorder, remove pages, compress, text notes, highlights, fill and sign.
  • Unlimited core workflows with no accounts, watermarks or AI features.

Limits: Compression is structural optimization only: it does not downsample images and may not shrink every PDF. Notes and highlights are flattened into the export. Signatures are visual electronic marks, not certificate-backed digital signatures, and do not verify identity.

Evidence: Live product, /privacy and /terms (effective 16 August 2026), checked 13 November 2026.

Space · Built for the NASA Space Apps 2026 junior astronaut challenge

Little Orbit

Live prototype

A junior astronaut mission trainer where students build a Moon or Mars outpost and balance power, oxygen, water, food and crew health over 14 turns.

Little Orbit mission control with resource gauges, a lunar outpost and today's flight plan.
Live site, captured 13 November 2026
Exact next-turn resource forecast and mission journal
Four scripted incidents, each with a free fallback
14 primary references: 13 NASA/JPL sources plus the challenge
ReactTypeScriptViteVitestPlaywright

AI

COS30049 Computing Technology Innovation Project

BuiltAI / full-stack / system integration proof-of-work

Chia's flagship university project: an AI, full-stack, and system integration mission that represents one of his strongest current proof-of-work projects.

Flagship university project and current favorite build
AI-oriented direction with full-stack delivery
System integration work across a real project repo

Currently the university project Chia is most satisfied with.

AIFull-stackSystem IntegrationSoftware Delivery

Negotiation · COS30018 Intelligent Systems, five-person team project

Velvet Hammer Auto Auction

BuiltAgent communication and protocols

A JADE multi-agent car auction in Java: a broker matches buyer and dealer agents, who negotiate prices automatically or manually from a Swing dashboard.

Broker, dealer and buyer agents exchanging JADE ACL messages
Three dealer strategies: stubborn, desperate and matcher
Live negotiation feed, market boards and a negotiation-gap chart

Team project led by Jonathan Zheng Li Chai. Chia's documented role is agent communication and protocols.

JavaJADEMulti-agent systemsJava Swing

AI · COS40007 AI for Engineering, Portfolio 4 (individual)

YOLOv5 Graffiti Detection

BuiltIndividual project

An individual deep-learning portfolio that trains YOLOv5 to detect graffiti, with annotation conversion, a dataset audit, three training iterations, IoU evaluation and error analysis.

Three iterations on 400, 600 and all 732 training images
Best default-configuration result: 18 of 40 test images reached IoU ≥ 0.90
Video detection, error analysis and a GenAI declaration

The target was 32 of 40 test images at IoU ≥ 0.90. The default-configuration notebook reached 18, which the write-up reports and discusses.

PythonYOLOv5PyTorchObject detectionJupyter

Cloud · Four-person team project, March to June 2026

Cloud File Management System

Portfolio proofTeam lead

A scalable, event-driven AWS architecture for uploading, processing and delivering files, covering storage, identity, queues, media processing and edge security.

Led a four-member team through the architecture design
Event-driven processing with Lambda, SNS and SQS
Edge and identity layers: CloudFront, AWS WAF, Cognito and Route 53

Architecture design project. There is no public repository, so this mission file is the public record.

Amazon S3CloudFrontAPI GatewayCognitoLambdaDynamoDBSNSSQSMediaConvertRekognitionAWS WAFVPC
Read case study

Problem

Design a file-management system that scales with uploads, processes media and images automatically, and stays secure at the edge.

Contribution and role

Chia led the four-member team that designed the architecture.

Architecture

  • Uploads land in Amazon S3 and are delivered through CloudFront, with Route 53 for DNS and AWS WAF at the edge.
  • API Gateway and Cognito handle the API and user identity. DynamoDB stores file metadata.
  • Lambda functions react to events, and SNS and SQS fan out and buffer work. MediaConvert transcodes video and Rekognition analyses images.
  • Private resources run inside VPC networking.

Technical challenges

  • Keeping processing asynchronous so uploads never wait for media jobs.
  • Choosing managed services that scale without servers to operate.

Results

  • A documented event-driven architecture across 13 AWS services.

Limits: This was a design project. Load, cost and availability figures were not measured in a live deployment, and no public source is linked.

Evidence: Chia's CV (September 2026). Completion of COS20019 Cloud Computing Architecture is recorded in his Degree Progress Audit (August 2026).

Database

COS20031 Database Design Project

BuiltCompleted unit

A completed university unit in relational database design: data modeling, normalization and SQL querying.

Unit completion recorded in the Degree Progress Audit (August 2026)
Database work is also applied in FinGence (Supabase and PostgreSQL)

Project artefacts are not public yet, so no project link is shown.

SQLDatabase DesignERDNormalization

Networking

TNE10006 Networks and Switching

BuiltCompleted unit

A completed university unit in network design and switching, covering VLANs, subnetting, STP and EtherChannel.

Unit completion recorded in the Degree Progress Audit (August 2026)
Followed by COS30015 IT Security; TNE30009 Network Security and Resilience is a current unit

Lab and portfolio artefacts are not public yet, so no project link is shown.

VLANsSubnettingSTPEtherChannel

skill system

Skills mapped to evidence.

ChiaOS avoids inflated ratings. Each skill is shown with a current stage and the work that supports it.

Clusters

6

Evidence items

18

Primary direction

AI

cluster 01

AI and Data

AI is the main direction: model workflows, experimentation, and practical product thinking.

cluster 02

Cybersecurity and Networks

Focused on secure foundations, network behavior, and systems that are practical to operate.

cluster 03

Full-stack / Web Systems

Building practical web systems that connect interface, logic, data, and deployment decisions.

cluster 04

Cloud and Architecture

Architecture thinking for systems that need clear boundaries, reliability, and safe deployment paths.

cluster 05

Database and Documentation

Turning project details into structured data, readable proof, and maintainable records.

cluster 06

AI-assisted Workflow

Using AI-assisted development to move faster while keeping proof, iteration, and clarity visible.

life os

Code. Learn. Build. Evolve.

The personal layer of ChiaOS: playful enough to keep building, serious enough to ship with proof.

Curiosity loop

Pick up a system, test the edges, and turn the lesson into visible proof.

Build loop

Use AI-assisted workflow to move faster while keeping the final result personal.

Refine loop

Cut noise, improve the interface, and make the next version feel more like ChiaOS.

career timeline

Growth logged as a system.

Each stage adds another layer: fundamentals, web systems, data, security, AI, cloud, and the operating system that now ties the story together.

01

stage

Foundation

Started with HTML, CSS, C, and programming basics.

SyntaxWeb foundationsProgramming logic

02

stage

Web Systems

Moved into structured websites, PHP, and database integration.

PHPDatabasesWeb workflows

03

stage

Data and AI

Explored KNIME, Python, model evaluation, and machine learning workflows.

PythonKNIMEModel evaluation

04

stage

Networking and Security

Built networking foundations around VLANs, subnetting, STP, and EtherChannel.

VLANsSubnettingSecure network design

05

stage

AI Computer Vision

Focused on real-time camera monitoring and evidence capture.

Computer visionIncident detectionEvidence

06

stage

Cloud Architecture

Designed AWS-based scalable and secure architecture concepts.

AWSScalabilitySecurity

07

stage

Career System

Created ChiaOS as a living career command center.

Portfolio systemProof of workCareer direction

build logs

The system keeps evolving.

November 2026

Mission archive refresh

Added Earth FM, FinGence, the bird classifier and the Sarawak tourism prototype as evidence-backed case studies with their stated limits, plus PagePilot and Little Orbit as public project entries.

ProjectsCase studiesEvidence

June 2026

ChiaOS v1 planning

Defined the personal career operating system concept, Original Setup route, Activated System route, local data model, and deployment path.

PortfolioPlanningNext.js

Now

Mission proof cleanup

Turning coursework and technical projects into clearer proof points for internship and long-term career positioning.

ProjectsProofCareer

Ongoing

AI-assisted builder workflow

Using AI tools to plan, code, review, and iterate faster while keeping the final site grounded in real local profile data.

AIWorkflowVibe coding

roadmap

Internship is a checkpoint, not the finish line.

Now

Build the operating system

Build ChiaOS, polish GitHub projects, prepare proof of work, and strengthen internship readiness.

Next

Enter real-world teams

Pursue internship exposure in software, AI, cybersecurity, or cloud environments.

Later

Specialize through practical systems

Grow toward AI engineering, cybersecurity specialization, and cloud-native secure systems.

Long term

Build intelligent secure systems

Create intelligent systems that are practical, secure, scalable, and useful beyond demos.

ask chiaos

Local answers from local data.

The assistant uses the profile, mission, skill, education, experience, build log, and roadmap data already inside this site.

Ask ChiaOS

Local profile assistant

Matched local data

I'm Chia, an AI-focused Computer Science student based in Kuching, Sarawak, Malaysia. I study Artificial Intelligence and Cybersecurity at Swinburne, and ChiaOS is my personal operating system for showing project proof, Realfun operations training, AI-assisted workflow, and what I am becoming.

Profile

signal handoff

Connect with Chia after the system run.

ChiaOS is built to show development taste, AI direction, operating proof, and career identity. The next signal can be feedback, collaboration, or a practical opportunity across AI and software systems.

AI-focusedProduct-mindedProof-ledOperations-tested