DS2010 — PERKESO Retraining Programme
Day 1 & Day 2 Reference Deck | Dr Caroline Ang, AI Growth Catalyst
Dr. Caroline Ang
Conscious Futurist · Growth Architect · ACIS
100K+
Global Community
3K+
Engagements
200+
Organisations
39 yrs
Practice
Doctorate UWTSD · MBA Cardiff Met · Knight's Award 2023
What I Do — And Why It Matters to You
My Work
AI & Intelligence Systems
Corporate Governance & Strategy
Executive Learning
Ventures & Ecosystem Building
Why I'm Here Today
To help you turn real experience into marketable, future-ready skills.
Download Files & Upload Your Work
Scan the QR code or visit the link below
linke.to/perkeso1
Slides & Activities
linke.to/perkeso
Shared Google Drive
Welcome to DS2010
You Are Already Further Along Than You Think
This programme is not about starting from zero. Every one of you brings real-world experience that data science tools will let you apply more powerfully.
Five-Day Course Map
Your DS2010 Learning Journey
Part 1
Introduction to Data Science
Part 2
Data Collection & Preparation
Part 3
Exploratory Data Analysis (EDA)
Part 4
Data Visualisation Techniques
Part 5
Introduction to Machine Learning
Programme Context
Who This Programme Is For
The Six of You
Six participants upskilling through PERKESO's Employment Insurance Scheme.
What You Will Walk Away With
- Structured problem-framing with evidence
- Practical data preparation experience
- AI tools for professional settings
Part 1: Introduction to Data Science
Part 1 — Core Idea
What Is Data Science, Really?
Data science is a disciplined way of turning raw information into evidence-based decisions.
Ask
Frame a clear, answerable question
Collect
Gather relevant, trustworthy data
Understand
Find patterns and test assumptions
Communicate
Share findings to enable action
How AI Changed Data Science Forever
Before AI
- Needed Python/R coding skills to analyse data
- EDA took days of manual work
- Only data scientists could do data science
After AI
- Ask questions in plain English — AI writes the code
- EDA in minutes with ChatGPT, Julius, or Manus
- Anyone with domain knowledge can contribute
AI did not make data scientists obsolete. It made data science accessible to everyone — including you.
Framework: Data to Decision
From Raw Facts to Real Change
Activity: Data, Information or Insight?
Classify These Three Statements
Classify each statement from the JobSeeker360 Dataset.
Use Case: Part 1
The Mystery of Slow
Re-employment
A programme manager at an employment-support organisation notices something troubling: some professionals in career transition find new work within weeks, while others are still unemployed months later. The organisation has data, but no one has yet asked the right question.
Use Case — Framing the Question
From Vague Concern to Clear Question
The first instinct is to ask "How do we help people get jobs?" That is too broad to answer with data. Good data science begins with a precise, measurable and responsible question.
"How might we use training, job-search activity, skills and support-channel data to understand which re-employment pathways are associated with shorter time to placement, without claiming that an association proves causation?"
Activity: Question Auction
Which Questions Are Worth Asking?
You have 10 tokens. Bid on the questions you believe are most useful and answerable.
Descriptive
"How many participants completed at least one training course?"
Diagnostic
"Why do participants in some skills categories take longer to find work?"
Predictive
"Which participants are most likely to be placed within 60 days?"
Prescriptive
"What support actions should the programme prioritise to reduce average placement time?"
The Data-Science Lifecycle
Eight Stages Every Project Should Follow
Define
Clarify the question
Collect
Gather appropriate data
Prepare
Clean and structure
Explore
Find patterns, anomalies
Model
Apply analytical techniques
Communicate
Present findings clearly
Act
Support a decision
Review
Evaluate and learn
Activity: Build the Pipeline
Assemble the Lifecycle in the Right Order
Arrange the eight lifecycle stages in order, then find the two mistake cards hidden in the deck.
Mistake Card 1
Building a model before checking the data. Errors, duplicates, or missing values will be amplified by the model rather than corrected.
Mistake Card 2
Assuming correlation proves cause. Two variables moving together does not mean one causes the other — always separate association from causation.
Understanding Data Types
Not All Data Is the Same
Numerical
Counts, measurements and amounts — such as days_to_placement or interviews_attended.
Categorical
Labels and groups such as skills_category, industry_background, or placement_status.
Date and Time
Dates, timestamps and durations that require consistent formatting to avoid errors.
Free Text
Unstructured open comments — rich in context but harder to analyse without careful handling.
The Human and AI Data Team
Who Does What in a Modern Data Project?
Human Roles
- Programme manager: owns decisions
- Data owner: interprets fields
- Analyst: prepares data
- Communicator: translates findings
AI Tool Roles
- ChatGPT: interactive analysis
- Custom GPT: repeatable assistance
- ChatGPT Project: persistent workspace
- Gemini Notebook: source-grounded research
- Manus: end-to-end analysis
Session Close
Your Problem Statement
Draft your personal problem statement using the template below.
The Question
"We want to understand [outcome] so that [decision-maker] can [action], using [data sources], without claiming [association] proves [causation]."
The Success Criterion
"We will know we have succeeded when we can show [evidence] to [audience] in a way that supports [responsible action]."
Part 2: Data Collection & Preparation
Session Opening
Revisiting Your Problem Statement
Before collecting or analysing any data, your problem statement must be clear enough to define exactly what data you need.
From Day 1
You identified a focused, responsible question about career transition pathways to guide today's work.
Day 2 Goal
Produce a data-readiness note certifying whether the synthetic dataset is ready for Exploratory Data Analysis.
The Fictional Dataset
Introducing the JobSeeker360 Dataset
Use Case: Part 2
The Messy Spreadsheet Rescue
The unvalidated JobSeeker360 Dataset contains six categories of problem that could distort any analysis built on top of it.
Duplicate Rows
The same participant appears more than once, inflating counts.
Inconsistent Labels
"IT" and "Information Technology" are treated as different categories.
Missing Values
Blank cells appear in placement_status and days_to_placement.
Mixed Date Formats
Some entries use DD/MM/YYYY; others use MM-DD-YY.
Invalid Numbers
Negative values appear in applications_submitted and days_to_placement.
Unclear Definitions
"Placement" is ambiguous — employed, or merely referred to an employer?
Activity: Data Quality Escape Room
Three Pairs, Three Missions
Three pairs, 20 minutes each. Find issues, document them, propose a fix.
Pair A — Missing & Duplicate
Find blank key fields and repeated participant IDs, then decide whether to remove or flag each duplicate.
Pair B — Labels & Formats
Spot inconsistent category labels and mixed date formats, then propose a single standard for each.
Pair C — Values & Definitions
Flag implausible numbers and ambiguous field names, then draft a clear definition for each unclear field.
The Data-Quality Wheel
Six Dimensions of Trustworthy Data
Activity: Keep, Question or Remove
What Should We Collect?
Not every field earns its place — decide what to keep, question, or remove.
Data Dictionaries
Why Every Field Needs a Clear Definition
A data dictionary is the contract between data creators and analysts — without it, the same field can be interpreted in entirely different ways.
Field Name
The exact column name as it appears in the file.
Description
A plain-language explanation of what the field means and how it was collected.
Data Type
Numerical, categorical, date, or free text — specifies how the field should be stored and handled.
Allowed Values
The complete list of valid entries or numerical range; anything outside is an error.
Owner and Source
Who is responsible for this field and where the data comes from.
Activity: The Data Dictionary Challenge
Definition Court
Five fields from the JobSeeker360 Dataset go on trial — four participants debate the correct definition before the group reaches a verdict.
The Analyst
Argues for the definition that makes the field most useful for analysis.
The Data Owner
Explains how the field was actually collected and what the original intent was.
The Programme Manager
Asks whether the definition supports the programme's reporting needs.
The AI Assistant
Offers a suggested definition based on common conventions — but the group must verify it.
AI Tool Demonstration
ChatGPT: Inspecting the Dataset
Upload the JobSeeker360 Dataset to ChatGPT Data Analysis and run a structured data quality inspection using this prompt:
"You are a data quality assistant. Please inspect this file and report: row count, column names, data types, missing values per column, duplicate rows, inconsistent category labels, suspicious numerical values, date-format issues and any fields whose meaning is unclear. Do not make recommendations or draw conclusions until I ask you to."
AI Tool Demonstration
ChatGPT Custom GPT: The Data Readiness Coach
The fictional Data Readiness Coach is a Custom GPT configured with four standing rules:
Always start with data quality
Never attempt analysis before reporting on the state of the data.
Explain issues in plain language
No jargon — any team member should understand the report.
Separate facts from hypotheses
Clearly label what is observed, inferred, or unknown.
Ask before changing anything irreversible
Confirm with the user before deleting rows, relabelling, or imputing values.
AI Tool Demonstration
ChatGPT Projects: Your Analysis Workspace
A ChatGPT Project is a persistent workspace where you can store files, chat history, custom instructions and saved outputs together.
- The synthetic JobSeeker360 Dataset (CSV)
- The data dictionary (document)
- Course notes from Days 1 and 2
- Your problem statement
- Saved outputs from previous sessions
AI Tool Demonstration
Gemini Notebook: Source-Grounded Learning
Gemini Notebook is a research workspace where you upload documents and ask questions that are answered using only those sources — with each claim cited back to its origin.
"Using only the uploaded sources, explain what each field in this data dictionary means, identify any missing definitions, and produce a short briefing for a beginner. Cite the source for each material statement. If the sources do not answer a question, say that the information is not available."
AI Tool Demonstration
Manus: End-to-End Analysis and Visual Output
Manus is an AI agent that handles end-to-end workflows: inspecting data, identifying patterns, generating visualisations and producing a presentation-ready brief from a single prompt.
Inspect
Audit the prepared synthetic dataset for quality issues.
Identify
Report no more than three evidence-based patterns, labelled as observations.
Visualise
Create one simple visual summary, such as placement status by support channel.
Explain
Write a plain-language explanation of the visual for a non-technical reader.
Brief
Produce a short decision brief separating observed results, interpretations and unknowns.
The AI Workbench
Choosing the Right Tool for Each Task
Responsible AI — A Practical Checklist
Eight Questions Before You Act on Any AI Output
Purpose
Is this AI use justified?
Privacy
Could this identify a person?
Data Quality
Has the data been validated?
Assumptions
What has the AI assumed?
Responsible AI — A Practical Checklist
Eight Questions Before You Act on Any AI Output
Human Review
Was output checked by a person?
Fairness
Could any group be disadvantaged?
Explainability
Can you explain it plainly?
Traceability
Is there a record of outputs?
Activity: The Data-Ready Handover
Is the Dataset Ready for EDA?
Prepare a data-readiness note before handing the dataset to the next session.
Activity: The Data-Ready Handover
Is the Dataset Ready for EDA?
Activity: Handover Relay
Passing the Baton to Day 3
As a group, complete the data-readiness note for JobSeeker360 and vote: Ready, Ready with conditions, or Not ready — more work needed.
If You Vote "Ready"
The dataset moves to EDA with the handover note as the official record.
If You Vote "Not Ready"
Identify the single most important issue, agree who owns it, and define what resolution looks like.
Preview: Part 3 — Exploratory Data Analysis
Part 3 Preview — EDA
What Is Happening in the Data?
Exploratory Data Analysis (EDA) systematically examines patterns, distributions, relationships, and anomalies in the cleaned dataset. EDA reveals what is happening — not why — and interpretation requires domain knowledge.
Distributions
How are values spread across a field, and how wide is the range?
Group Comparisons
Do placement rates differ by support channel or skills category?
Outliers and Anomalies
Which data points fall far outside the typical range, and are they errors or genuine exceptions?
Preview: Part 4 — Data Visualisation Techniques
Part 4 Preview — Visualisation
Turning a Finding Into a Chart
Day 4 covers the principles of effective data visualisation: choosing the right chart type, designing for clarity, and avoiding common misrepresentations. As a preview, consider how one validated EDA finding might be communicated visually.
Chapter 5
Preview: Part 5 — Introduction to Machine Learning
Part 5 Preview — Machine Learning
Supporting Decisions Without Replacing Judgement
Feature
Input variable the model uses.
Label
Outcome the model is trained to predict.
Prediction
Model's output for a new participant.
Five-Day Handover
Who Teaches What — and When
Reference Materials
Reusable tools for post-class practice and independent study.
Prompts for ChatGPT, Manus and Gemini Notebook
ChatGPT — Data Inspection
"Inspect this dataset and report: row count, column names, data types, missing values, duplicate rows, inconsistent labels, suspicious values and unclear fields. Do not recommend actions yet."
Manus — End-to-End Brief
"Inspect the attached data, identify up to three evidence-based patterns, create a visual summary of [finding], and produce a short decision brief separating results, interpretations and unknowns."
Prompts for ChatGPT, Manus and Gemini Notebook
Gemini Notebook — Grounded Briefing
"Using only the uploaded sources, explain each field in this data dictionary and produce a short briefing for a beginner. Cite sources for every material statement; say 'not available' if a question isn't answered."
Any Tool — Responsible Framing
"Separate your response into: (1) what the data clearly shows, (2) what might be inferred but needs investigation, and (3) what is unknown. Do not mix observations with interpretations."
Before You Pass Any Dataset Forward
- All rows with the same participant_id represent the same, non-duplicated record
- No key fields (outcome, identifier, date) contain blank or null values without documented justification
- All categorical fields use a single, consistent label list with no spelling variants
- All date fields use a single agreed format throughout the entire file
- Every field has an entry in the data dictionary with a clear, agreed definition
- Fields containing potentially sensitive information have been flagged, reviewed, and the source and version of the file are recorded and traceable
A Field Definition You Can Reuse
Reference: Data-Readiness Handover Template
Your Professional Handover Note
Purpose
Question this dataset was prepared to answer.
Dataset Version
File name, date, and version number.
Changes Made
Cleaning actions: duplicates, labels, formats corrected.
Limitations
What the dataset cannot answer or covers.
Privacy Considerations
Sensitive fields flagged and how handled.
Reference: Participant Reflection
What Have You Learned? What Will You Do Next?
What I Understood
Which concept gave you the clearest "aha" moment?
What I Found Difficult
Where did you feel uncertain or confused?
What I Will Try
Name one concrete action you will take in the next two weeks.
Looking Ahead
What Happens in the Sessions Ahead
Day 3 — Exploratory Data Analysis
Patterns, distributions, and outliers revealed.
Day 4 — Data Visualisation Techniques
Right chart, clear insight communication.
Day 5 — Introduction to Machine Learning
Build your first classification model.
Meet Your Cohort
Six Professionals, Six Perspectives
Professionals from diverse fields are upskilling together through PERKESO's Employment Insurance Scheme.
Your Backgrounds Are an Asset
Content Moderation
Understanding harmful patterns and human judgement at scale is directly relevant to data labelling and model evaluation.
3D Animation
Expertise in rendering pipelines is directly relevant to data visualisation and communicating complex findings.
Your Industry
Your specific domain knowledge helps you ask better questions and find more relevant insights than a pure technician could.
Your AI Toolkit — Live Demos
Free tools available now — try them today
ChatGPT Data Analysis — Upload & Interrogate
Free at chatgpt.com — no account required for basic use
Upload Your Data
Click "+" and upload your CSV file.
Summarize the Dataset
Ask ChatGPT to describe rows, columns, and fields.
Identify Key Trends
Ask which sector has the highest time-to-employment.
Visualize Data
Request a bar chart by age group.
Export Your Insights
Download the generated chart as an image.
Note: ChatGPT writes Python behind the scenes — you just ask.
Gemini Notebook — Source-Grounded Research
Free at gemininotebook.google.com — sign in with any Google account
What You Can Upload (up to 50 sources per notebook — free)
PDFs
Research papers, policy documents
CSVs & Spreadsheets
Sheets and Excel files
Google Docs & Word
Docs and Word files
YouTube & Web Links
Videos and web pages
Gemini Notebook — Live Demo Steps
Follow along at gemininotebook.google.com
Go to gemininotebook.google.com → Create new notebook
Upload: the synthetic JobSeeker360 Dataset + any PERKESO policy PDF
Ask: "What are the main factors affecting re-employment speed according to these sources?"
Ask: "Generate a study guide summarising the key variables in this dataset"
Click "Audio Overview" to generate a podcast-style summary you can listen to
Note: Each answer includes a citation — click it to see exactly which source it came from.
Claude — Reading PDFs and Writing Reports
Free at claude.ai — sign in with email
Upload a PDF Report
Extract key statistics or summarise any uploaded document.
Compare Two Documents
Ask Claude what changed between two document versions.
Write a Data Brief
Turn EDA findings into a one-page executive brief.
Check Your Analysis
Verify that your conclusion logically follows from the data.
Note: Claude can also output Excel, Word, and PowerPoint files.
Manus — Autonomous End-to-End Analysis
Free tier at manus.im — 300 credits refreshed daily
Sign up at https://manus.im/invitation/SYJJCJGUFWXF
Give it a focused task
Manus writes code and charts
Download the finished report
Gamma — Turning Your Analysis Into a Report
You are already using it — here is what else it can do
- Upload findings → Gamma structures it visually
- Add charts, infographics, or summaries instantly
- Share a live link or export as PDF
Gamma — Live Demo: Build a Mini Report
Follow along at gamma.app
Open a new Gamma document
Paste: Here are my EDA findings: [paste 3 bullet points from your analysis]
Ask: Turn this into a 3-section report with a summary, a chart, and a recommendation
Share the link with your trainer
Your LinkedIn Profile Is a Dataset
And most people never read it.
Profile Data
Your headline, summary, skills, and experience are structured, indexed, searchable fields.
Engagement Data
Post impressions, reactions, and shares reveal what content resonates with your audience.
Audience Data
Profile viewers' industries and job titles give you free market research about who's noticing you.
Live Demo — Dr. Caroline Ang's LinkedIn Dashboard
linkedin.com/in/drcarolineang
Look at the four key metrics on screen. What story do you think the data is telling?
What Do These Numbers Actually Mean?
846 Post Impressions (7 days) ↑299%
One post dramatically outperformed — find it and repeat it.
6,349 Total Followers ↑1%
Steady growth — but are they the right audience?
904 Profile Viewers (90 days) ↑174%
Something triggered a surge — a post, comment, or share.
61 Search Appearances (Aug 4–10) — 0% change
Keywords work but aren't expanding — update your headline.
Use AI to Analyse & Improve Your LinkedIn Profile
ChatGPT + Claude as your personal career strategist
Profile Audit
"Score my LinkedIn headline and summary out of 10. What are the top 3 things to change?"
Keyword Gap Analysis
"What keywords am I missing that recruiters would search for in this role?"
Rewrite My Headline
"Write 5 LinkedIn headlines for someone upskilling in data science. Make them searchable and human."
Interpret My Analytics
"What story do these LinkedIn metrics tell? What should I do differently this week?"
Your Resume vs The Job Market — Who's Winning?
Job Market Analysis
Analyse 10–20 job listings to extract the most common skills, keywords, and qualifications recruiters actually want.
Resume Gap Analysis
Compare your resume against those listings to identify exactly what you're missing and what to learn next.
Career Coach Simulation
Feed your resume and a job description to ChatGPT or Claude and ask it to score your fit and build a development plan.
Step 1 — Analyse the Job Market with AI
What does the market actually want from a data analyst in Malaysia?
Collect 10 Job Listings
Search "Data Analyst Malaysia" on LinkedIn, JobStreet, or Indeed and copy 10 full job descriptions into one document.
Ask ChatGPT to Extract the Patterns
Paste all 10 descriptions and ask ChatGPT to list the top 10 most frequently required skills, tools, and qualifications with their frequency.
Build Your Skills Frequency Table
ChatGPT returns a ranked list — common findings include Excel (9/10), SQL (8/10), Power BI (7/10), and communication skills (10/10).
Identify Must-Haves vs Nice-to-Haves
Ask which skills appear in more than 70% of listings to get a clear, prioritised learning roadmap.
Step 2 — Run Your Resume Through AI
Gap analysis, fit score, and a personalised development plan — in 5 minutes.
The Resume Gap Analysis
"Score my fit out of 100. List my strengths, gaps to close, top 3 skills to learn next, and one sentence to add to my LinkedIn summary."The Career Coach Simulation
"Ask me the 5 most likely interview questions for this role. After each answer, give me recruiter feedback."This is not cheating — this is data-driven career strategy. Every competitive job seeker should be doing this.
Activity — Analyse Your Own Job Target
15 minutes. Real job. Real resume. Real feedback.
I Have a Target Role
Paste a real job listing and resume into ChatGPT.
I'm Still Exploring
Ask ChatGPT to find top skills across 5 job listings.
I Want Interview Practice
Ask Claude what role suits you, then practise interview questions.
Choosing the Right Tool for the Job
Try It Now — 20-Minute AI Sprint
Pick one tool, one question, one output
The Analyst (ChatGPT)
Upload the CSV and screenshot your best chart.
The Researcher (Gemini Notebook)
Upload CSV, ask about career transitions, generate Audio Overview.
The Writer (Claude)
Paste findings and ask Claude for a 200-word summary.
Debrief: What surprised you?
Turn Your New Skills Into Freelance Income
Fiverr (fiverr.com)
- Create gigs starting from RM20–RM50
- Popular: data cleaning, Excel dashboards, charts
- Start specific: "I will clean your spreadsheet."
Freelancer (freelancer.com)
- Bid on projects from clients worldwide
- Popular: data entry, analysis, report writing
- Filter by Fixed Price projects under RM500
Your first freelance project does not need to be perfect — it needs to exist. Every expert started with their first RM50 gig.
AI Tools Don't Replace Your Thinking — They Amplify It
You Are the Analyst
You define the question and make the decision.
Start Small, Build Confidence
Pick one tool. Try one task. Repeat.
Your Background Is Your Edge
Your prior career is your competitive advantage.
Connect with Me
Data Science Foundation