#AIThursday · free live session

Build a second brain
for your company

A knowledge base your team and your AI can both read. Obsidian and Claude, shown end to end.

Thursday 24 September 2026 · 9:00 to 10:30 PM · Google Meet

Azhari Razak · AI Systems Consultant
Scan for the Claude class
aitraining2u.com
Your trainersAITraining2U

Led by practitioners, not theorists

Chan Wei Khjan

Chan Wei Khjan

Audit Partner, YYC · MIA Board

Chartered Accountant (ACCA · C.A.(M) · FCA Singapore) and MIA board member. Featured in Business Insider for pioneering AI inside the accounting profession.

Marcus Chia

Marcus Chia

AI Product Manager & Harness Engineering Expert

Builds and ships AI workflows in the wild — Claude projects, MCP servers, multi-agent pipelines and data dashboards. Turns messy business problems into working AI solutions.

Warren Leow

Warren Leow

Founder, AITraining2U PLT

Drives AITraining2U's mission to equip 100,000 Malaysians with practical AI skills — hands-on with AI agents, automation and applied analytics for enterprise teams.

TonightAzhari A. Razak

Azhari A. Razak

Multi-Agent Systems Builder

Founder of Urbanite Enterprise, an AI consultancy for Malaysian SMEs. Nineteen years of mission-critical network operations at Telekom Malaysia, now delivering Claude, n8n and multi-agent workflows in production.

Azhari leads tonight's session. The same team teaches every AITraining2U class.
Our clients1,800+ professionals trained

Trusted by leading Malaysian organisations

Banking, telco, retail, healthcare, energy and government-linked companies. Teams across Malaysia upskill with AITraining2U.
TonightWhat this is

A show and tell, and I will be honest about it

I am going to show you a knowledge base that is actually running, tell you what it does for me, and point at how it is built.

✓

You will leave knowing

What a second brain is, what it buys you, and where to start on Monday.

✗

You will not leave with

A step by step build. That is a two day class, and ninety minutes would do it badly.

◎ Why say this first

Because the worst thing a session like this can do is promise a workshop and deliver a trailer. You should know which one you are in.

Ask anything as we go. The questions are usually better than the slides.
Who is talkingNineteen years in telco

From the cable, to the contract, to the code

2006

Engineer

Electrical and electronic engineering, Imperial College London. Straight into Telekom Malaysia.

2006 to 2015

Submarine cables

Eight years restoring international cables, ending as the manager of the restoration office.

2015 to 2019

Quality and control

Head of a quality unit through an ISO 9001 migration, then business control for the network operations centre.

2019 to 2025

Enterprise sales

Six years as an account manager for education customers, public universities and the ministry.

2026

AI Systems Consultant

Left on a voluntary separation plan. Started again, with a laptop and one folder.

Three careers inside one company. None of them was in AI.
Who is talkingThe telco years

Nineteen years of keeping the lights on

Appreciation Night
November 2014

Appreciation Night

Back then I ran the office that restored all of TM's international cables. Best Manager, 2013.

The operations floor
April 2015

The operations floor

Where the network is watched around the clock, and where nothing works unless the last person wrote it down.

A career spent on things that must not break. That is the mindset I brought to AI.
Who is talkingWhat carried over

What the cable years taught me

When a cable breaks
Restoration
  • You do not improvise an outage. You follow the plan written after the last one, and you update it after this one
When the auditor comes
Quality
  • ISO 9001 is one idea: write down how you work, then prove you did it. That is a second brain with a certificate
When the account moves
Sales
  • An account outlives the people on it. What the customer was promised has to survive every handover, or it is promised again
☞

The thread

Every job I did well depended on something someone wrote down. AI did not change that lesson. It made it pay ten times over.

2026The student

I sat where you are sitting

Claude
19 June 2026

Claude

AI agents as digital colleagues. My first AITraining2U class, and I was asked to show the room my own system.

Vibe coding
28 June 2026

Vibe coding

Rapid prototyping. Building working apps by describing them, nine days later.

n8n
17 July 2026

n8n

Agentic automation. Wiring assistants into the tools a business already runs on.

Three classes in four weeks, then a five day certification. All of it as a participant.
2026The teacher

Five weeks later, I was teaching it

The Claude class
22 July 2026

The Claude class

Invited to help teach the two day class, the biggest one yet. I taught the knowledge base and multi agent section.

Day two, my section
21 August 2026

Day two, my section

The same material I am showing you tonight, taught hands on with the room building along.

And again
27 August 2026

And again

The knowledge base and multi agent day, now on the class's trainer roster.

Student in June. Teacher by the end of July. The difference was one folder, kept every day.
Show and tellMy favourite build

One person, one assistant, a whole team

Infographic: one person, one orchestrator, twenty specialist roles
1

One front door

I talk to one assistant. It decides which specialist the work belongs to and briefs them.

2

Twenty written roles

Researcher, archivist, writer, reviewer and more. Each one is a note describing a job, with its own memory.

3

One second brain underneath

Every role reads from and writes to the same knowledge base. Without it, they are twenty strangers.

◎ Why it is my favourite

It is the clearest proof of tonight's point. The team is only as good as what is written down.

Show and tellThe first build

Nineteen years of work, found in seconds

Infographic: a private search system over a 19 year career archive, run by a team of generic AI helper robots
1

My own archive

Nineteen years of documents and correspondence, searchable by meaning, not exact words.

2

An ordinary laptop

Open source tools, 8GB of memory, and nothing leaves the machine.

◎ How this was made

Redrawn with AI from my June original. The numbers are June's. It has grown since.

Show and tellA research engine

A library that answers back

NotebookLM infographic: EDITH, a research to deliverable engine
1

238 expert videos

Read in full and queried as one library. Every answer cites its source.

2

Answers become assets

Explainers, slides and reports, made from the same grounded answers.

⚠ Look at the small print

NotebookLM made this in one pass, and some of the small text is garbled. I left it that way on purpose. It is the best argument for checking I have.

The problemThree questions

Three questions a folder cannot answer

1
Audit
  • What did the auditor query last year, and how did we answer it?
2
Commercial
  • What terms did we agree with that vendor, and who agreed them?
3
Decisions
  • Why did we decide it that way, and who was in the room?
☞

The real test

You have those answers somewhere. The question is whether you can get them back in under a minute, and whether they survive the person who knew them leaving.

The costWhy it matters

When someone resigns, it walks out the door

●

It is not lost, it is unfindable

The file exists. Nobody knows which folder, which version, or which of the three copies is current.

●

So the work gets redone

The same vendor gets re-evaluated. The same policy gets re-written. Nobody calls it rework because nobody counts it.

●

And the answer cannot be checked

Somebody remembers the number. Nobody can show where it came from.

shared drive
Final_v2.docx
Final_v2_REVISED.docx
Final_v2_REVISED_use_this_one.docx
notes.txt
scratch.txt
meeting (1).docx
meeting (1) copy.docx

Everyone recognises this folder. That recognition is the whole problem.

A
Part one

What a second
brain actually is

Not an app you buy. A folder, a habit, and two rules.

AITraining2U #AIThursday · 24 September 2026 · Azhari Razak
The ideaObsidian

A vault is a folder. That is the whole trick.

Lateral Logistics(vendor)Q3 Targets[[ ]]Project Atlas[[ ]]Contract, signedMeeting, 12 Sept
vendor-note.md
---
type: vendor
tags: [procurement, freight]
---

# Lateral Logistics

Offer: RM11.80 per unit, 12 month
commit, NET30.
Risk: only two depots.

Ties into [[Q3-Targets]]
Blocks  [[Project-Atlas]]

Plain text on your own disk. No database, no lock in.

If you can write a memo, you can write one of these. That is the entire skill.
At scaleMine, after a year

This is the system behind tonight

A dense point cloud, one point per note

Ninety five thousand notes. I am not going to open it, and you would not want me to.

◎ What you are looking at

One point per note. It started as one folder and one habit, and nothing in it was planned.

✓ The part that matters

I can still find anything in it in seconds, and so can the assistant. That is not because it is tidy. It is because it is linked.

How it worksFour moves

Four moves. Everything else is detail.

1. One noteper thing2. Link morethan you folder3. An indexat the root4. A maintenancehabitevery week,it gets smarter
1

One note per thing

Person, project, vendor, decision. One canonical note each, never three copies.

2

Link more than you folder

A note belongs to many topics. A folder only lets it belong to one.

3

An index at the root

So the assistant reads a map and follows two links.

4

A maintenance habit

One note after each meeting, a weekly tidy. The move that compounds.

I am naming these rather than walking them. Walking them with someone watching is the class.
Why structure paysThe honest version

Small pile, no difference. Big pile, all the difference.

If you have seven notes, structure buys you nothing. It starts paying at a few hundred, sooner than people expect.

Without a map: read everythingQuestion500 notes, all readslow, costlyWith an index: read the map, follow two linksQuestionIndexnote Anote Bfast, cited
◎ Three things you get

Faster, because less is read. Cheaper, for the same reason. And more accurate, because the few notes it reads are the right ones.

✓ And the one that matters most

You can check the answer. Which is the next thing I am going to show you.

The jargon, decodedRAG

RAG is just an open book exam for the AI

Retrieval Augmented Generation. Scary name, simple idea: before the AI answers, it looks things up in your notes first, the way you would check a file before replying to your boss.

1 · Retrieve
Find the right pages
  • You ask a question. It searches your vault and pulls out the few notes that matter
  • Like a librarian fetching three files, not the whole cabinet
2 · Augment
Put them on the desk
  • Those notes are handed to the AI together with your question
  • Now it is reading your facts, not guessing from memory
3 · Generate
Write the answer
  • It writes a reply based on what it just read
  • And it can point to the exact note it used, so you can check
☞

Why your second brain makes RAG work

Without it, the AI is sitting a closed book exam: it answers from general knowledge and fills gaps with confident guesses. A tidy, linked vault is the book. Good notes in, good answers out.

B
Part two · live

Let me show you
a real one

A working vault, opened on screen. Everything in it is invented, so I can click anywhere.

AITraining2U #AIThursday · 24 September 2026 · Azhari Razak
LiveWhat to watch for

Three things, then I ask it a real question

The graph
Thirteen hundred notes
  • Nobody drew this. It is what the links look like from above
The folders
Work, clients, personal
  • Personal material sits apart, and that separation is a folder, not a setting
Three kinds of file
Instructions, roles, memory
  • That is the entire architecture. There is no fourth thing
☞

Why a copy and not the real one

The privacy control is which folder you point the assistant at, not which assistant you use. So tonight I point it at a copy where every client, project and person is fictional.

LiveThe question

An answer you can check

◎ What I am asking it

What did we agree with the cold chain client, and what has to happen before it is final?

●

The answer is not in one note

The terms sit in one, the approval in another, the open action in a third. It has to follow the links.

●

And it names its sources

Not "according to your files". The actual note, which I can open in front of you.

grounded answer
Agreed: the rate and terms in the
scope letter, subject to two things.

Before it is final:
 1. the signed variation
 2. finance clearing the cash impact

Sources:
  [[Tanjung-Cold-Chain]]
  [[Decisions-Log]]

Illustrative shape. The live one is run on screen.

This is the difference between an answer and a source. One you can act on. One you can only hope about.
The pointTrust

Anyone can get an answer. Few can check one.

✓

A colleague can verify it

You forward the answer and the note. They do not have to trust you, and you do not have to remember.

✓

It says when it does not know

Ask it something the vault does not contain and it should tell you so, rather than inventing something plausible.

✓

And it survives you

The next person asks the same question and gets the same answer, with the same source.

⚠ The honest caveat

None of this makes the assistant right. It makes the assistant checkable, which is a different and more useful property.

In anything that gets signed, a person still approves. The system does not remove that, and it should not.

Being right is the model's job. Being checkable is the system's job.
Be realisticWhen to bother

Worth doing, and when it is not

✓ Worth it
You will feel it within weeks
  • Knowledge that outlives one person: decisions, policies, vendor terms
  • Questions that get asked again and again by different people
  • Anything where being able to show the source matters
  • A team where the expert is a bottleneck
⚠ Not worth it
Do not start here
  • A one off project that ends next month
  • Material that is already structured elsewhere and maintained
  • A team of one who genuinely remembers everything
  • Anything you would not open again
☞

Where to start on Monday

One folder. One note after your next meeting. If you do that for a month and never open it again, you have lost an hour and learned something true.

C
Part three

What becomes
possible after

Once the knowledge is written down, it stops being a filing exercise.

AITraining2U #AIThursday · 24 September 2026 · Azhari Razak
By department1 of 4 · Finance

Finance: the numbers, and why they are the numbers

What it holds
A note per thing
  • Vendor terms and payment schedules
  • Budget decisions, and who approved them
  • Audit queries, and how each was answered
Questions it answers
Each with a source you can open
  • What did the auditor query last year, and how did we answer it?
  • Which vendors are on NET30, and who agreed it?
  • Why was the Q3 budget cut, and who signed it off?
What an agent can do
Because it can read all of it
  • Prepares the month end pack, every figure linked to its source note
  • Flags a payment term that contradicts the signed contract
Examples drafted with Claude, names from tonight's fictional vault. Swap in your own department.
By department2 of 4 · Marketing

Marketing: stop relearning what already worked

What it holds
A note per thing
  • Every campaign brief, and what it actually delivered
  • Brand rules and the claims you are allowed to make
  • What customers said, in their own words
Questions it answers
Each with a source you can open
  • Which campaign brought the best leads last year, and why?
  • What are we allowed to claim about the product?
  • What did we try in Q2 that we should not repeat?
What an agent can do
Because it can read all of it
  • Drafts a new brief that starts from what worked, citing past campaigns
  • Checks copy against the approved claims before it goes out
Examples drafted with Claude, names from tonight's fictional vault. Swap in your own department.
By department3 of 4 · Sales

Sales: the account outlives the account manager

What it holds
A note per thing
  • One note per account: contacts, history, terms agreed
  • Objections heard, and how they were answered
  • Pricing decisions and their approvals
Questions it answers
Each with a source you can open
  • What did we promise Tanjung Cold Chain, and is it signed?
  • Who else asked for this discount, and what did we decide?
  • What does a new account manager need before the handover?
What an agent can do
Because it can read all of it
  • Writes a one page brief before each call, from the account and its links
  • Turns meeting notes into follow ups, and updates the account note
Examples drafted with Claude, names from tonight's fictional vault. Swap in your own department.
By department4 of 4 · Projects

Projects: decisions with their reasons attached

What it holds
A note per thing
  • Decisions, and the reasons behind them
  • Risks, owners and dates
  • Lessons from the last project
Questions it answers
Each with a source you can open
  • Why did Project Atlas change scope in August?
  • Which risks are open, and who owns them?
  • What went wrong the last time we used this vendor?
What an agent can do
Because it can read all of it
  • Writes the weekly status report from the decision log, with sources
  • Warns when a new decision contradicts an earlier one
Examples drafted with Claude, names from tonight's fictional vault. Swap in your own department.
Putting it togetherAgents and connections

One second brain, many agents

The vaultone folder,linked notesFinance notesbudgets, auditMarketing notescampaigns, resultsSales notesaccounts, termsProject notesdecisions, risksOne client note,shared by bothFinance agentContent agentSales agentProject agent
1

Every agent reads the same notes

So the finance agent and the project agent never disagree about what was decided.

2

Links carry knowledge across teams

One client note serves sales and projects. Nobody keeps a second copy.

◎ Live, if we have time

Tell me your department, and I will ask Claude for use cases, against tonight's fictional vault.

Before the filmNotice

The Lantern

An original short film, eighty seconds, made with AI tools for this session.

◎

Every frame was planned first

Fifteen shots, each written down before anything was generated.

◎

The company in the story does not exist

It is a fictitious Malaysian solar installer, invented for the film.

⚠ Trademark notice

n8n is a trademark of n8n GmbH. This film is an independent work and is not affiliated with, endorsed by or sponsored by n8n GmbH.

Any other names, marks or brands appearing in it remain the property of their respective owners.

Show and tellEighty seconds

A short film, made the same way

This is not a knowledge base. It is what the same habit of writing things down lets a small team do next.

●

The story

A solar installer makes one wish. A team of assistants picks it up, hires the specialists it needs, and hands back something that works.

●

Why show it here

Every character in it is a written role. That is the same idea as a note: if you can write the brief, you can build the thing.

⚠ Before you ask

Yes it is a bit rough in places. It was made on a small budget of credits, by the team, as a piece of our own work rather than a client job.

The company in it does not exist.

Watch what each character is actually doing. Every one of them is a written job description.
That wasHow it was made

Written down first, then made

1

A storyboard before a single frame

Fifteen shots, each with its own note: what happens, what is said, what it must not do.

2

Every stage approved before it ran

Nothing was generated until someone agreed what it would cost and what it was for.

3

And a reviewer before it was shown

Which caught things the people who made it could not see any more.

◎ Trademark and production notice

The film is an original work made for this session. Any names, marks or brands appearing in it are the property of their respective owners, and the company shown in the story is fictitious.

Same lesson as the vault. Write it down, and the thing becomes possible.
HonestlyWhat it takes

None of this is hard. It is just unfamiliar.

Tonight
Ten minutes
  • Install Obsidian, make one folder, write one note about something you decided this week
This month
One habit
  • One note after each meeting. Link the people and the projects you mention
The part I skipped
The actual building
  • The prompts, the structure, the maintenance, and how to point an assistant at it safely
☞

I was told to give you some, but not all of it

That was a fair instruction and I have kept to it. What I skipped is a two day class where you build it in the room, with your own files, and leave with it working.

Keep buildingAITraining2U.com

Our courses at AITraining2U

Public class

AI Agentic Automation with n8n

For ops, finance and non technical teams drowning in repetitive work. Build no code agents for chatbots, reconciliation, OCR and reporting.

15 to 16 Oct · TTDI
12 to 13 Nov · Bandar Utama
Third cohort

AI Agentic Automation with Copilot365

For Microsoft 365 organisations and IT governed teams. Governed agents across Word, Excel, Teams and SharePoint.

5 to 6 Nov · TTDI · 20 seats
Flagship

Mastering Claude and Multi-Agent Orchestration

For power users and teams standardising on Claude. Claude Code, Cowork and multi agent systems mapped to your business.

8 to 9 Oct · KL Eco City
22 to 23 Oct · TTDI
19 to 20 Nov · TTDI
Public class

AI Vibe Coding and Rapid Prototyping

For non technical founders and product managers. Build and ship working apps with Claude Code and vibe coding tools.

1 to 2 Oct · Subang
Technical track

AI Engineering

For developers, MLOps and data engineers. Production AI with Google ADK, RAG, MCP and agents, model agnostic.

11 to 13 Nov · KL Eco City
Book your seat

Scan to register

Opens the booking form. Pick a course and a date.

All courses are HRDC claimable, as enterprise or public workshops. AITraining2U.com
✓
Next

Come and build
one properly

Mastering Claude and Multi-Agent Orchestration, by AITraining2U. Two days, hands on, HRD Corp claimable.

Details and dates from Warren, right after this.

Thank you for staying up. Questions are welcome. · Azhari Razak