All resources
    📖Stories· Feature

    The Minister Who Codes

    How Singapore's Foreign Minister built a personal AI assistant — and published the blueprint for everyone.

    Dr. Vivian Balakrishnan · Singapore Minister for Foreign Affairs
    May 18, 2026
    8 min read
    · AI Strategy· Leadership· Open Source
    Dr. Vivian Balakrishnan on stage at AI Engineer Singapore 2026
    Slide: A second brain for a diplomat — Vivian explains his architecture
    Slide: Security is critical — Vivian on the bigger picture
    AI Engineer SG 2026
    "The alpha is in deployment. If we can't lead in model development, we can lead in how we use it."
    Dr. Vivian Balakrishnan, Singapore's Foreign Minister
    Dr. Vivian Balakrishnan on stage at AI Engineer Singapore 2026 — 'I built a personal AI assistant on a Raspberry Pi'
    Dr. Vivian Balakrishnan on the keynote stage at AI Engineer Singapore 2026.
    AI Engineer Singapore 2026

    🤔 Picture this for a moment

    A Foreign Minister — the kind of person we usually imagine busy with diplomatic affairs, sharp suits, schedules packed with heads of state — in his spare time, learning to code and building a personal AI assistant to help with his job.

    Then, instead of keeping it to himself, he published the blueprint publicly on the internet — so others could learn and build their own version.

    This isn't fiction. This is what Dr. Vivian Balakrishnan, Singapore's Foreign Minister, did at the AI Engineer Singapore 2026 conference.

    💡 His core message, in plain language

    Vivian didn't talk about "large language models" or "transformer architecture". His message is far simpler — and that's exactly why it's powerful.

    "The tools already exist. The question is: are you going to use them or not?"

    📖 Four big ideas to take home

    1. 🌱 AI value lives at ground level, not at the top

    Many people talk about AI as a high-level policy matter — this government, that corporation, those regulations. Vivian says that's not where real value is created.

    AI value is created when:

    • 👩‍⚕️ A doctor uses AI to read lab results faster
    • 👨‍🏫 A teacher uses AI to create more personalized materials
    • ⚖️ A lawyer uses AI to study thousands of pages of contracts
    • 👨‍💼 Someone like you uses AI for your daily work

    2. 🧰 Stop waiting for the "perfect" tools

    Many people say: "I'll wait until AI gets better." Or: "I'll wait until my office provides it." Vivian: no need to wait. The tools that exist today are already enough to change how you work.

    3. 🔗 You don't need to build — you just need to assemble

    Imagine AI tools like ingredients in a kitchen. You don't need to grow rice, milk cows, or grind flour yourself. You just need to cook them into a dish that suits you.

    Vivian calls this "writing the glue" — writing the "glue" that connects existing tools into something useful for you.

    4. 🏋️ The only way to understand AI: use it

    "You can't understand technology just by looking at it. You have to do it."

    Vivian himself — a busy minister — makes time to actually code. Reading about AI is like reading about swimming. You'll never learn unless you get in the water.

    His line of thinking
    From the problem to the answer
    1
    The problem
    AI top-down policy isn't where the macro story is. Real value gets created on the ground level.
    2
    Who benefits
    Ordinary people using tools — doctors, lawyers, teachers — in their day-to-day work.
    3
    Make ourselves better
    Use the tools to upgrade how you actually do your job.
    4
    Re-engineer your life
    Don't just patch — rethink whole workflows around what AI can now do.
    5
    How do we actually do it?
    Vivian's answer comes in three parts.
    A
    Tools already exist
    It's about IF and HOW you use them.
    B
    Write the glue
    Don't build new tools — connect the ones that exist.
    C
    Learn by doing
    You can't understand tech just by looking — you have to use it.
    The alpha is in deployment
    If we can't lead in model development, we can lead in how we use it.
    Reconstructed from the AI Engineer Singapore 2026 keynote
    Vivian Balakrishnan presenting his 'Security is critical' slide
    "A static approach will not work — end users need to be part of the conversation."
    AI Engineer Singapore 2026

    🛠️ What tools does Vivian use?

    Here's the most interesting part: Vivian published his complete recipe openly on GitHub. Anyone can see, learn, even copy it. He calls it "NanoClaw — Second Brain for a Diplomat".

    Simple analogy: an AI assistant is like a new employee who needs a brain, a manager, memory, ears, an entry door, and a filing cabinet.

    Slide 'A second brain for a diplomat' — Vivian's full architecture
    Vivian's own slide: the blueprint he published for everyone.
    AI Engineer Singapore 2026
    Architecture diagram

    How all the tools wire together

    User → Cloud
    01
    Interfaces
    01.01
    Nura
    User
    01.02
    Telegram
    Chat
    01.03
    iPhone
    Obsidian
    01.04
    Mac
    Obsidian
    02
    Brain & Memory
    02.01
    Mnemon
    Memory graph
    02.02
    Mas nano
    Claude in Docker
    02.03
    OneCLI
    Vault key
    03
    Model & Storage
    03.01
    Anthropic
    Claude API
    03.02
    Vault
    Notes folder
    UserLocal toolMemory / dataKey / accessExternal API
    Source: NanoClaw blueprint by Dr. Vivian Balakrishnan
    🧠

    Claude

    The brain

    AI from Anthropic, similar to ChatGPT but from a different company. Handles all the "thinking" — answering questions, writing, summarizing.

    View on GitHub
    👔

    NanoClaw

    The manager

    Open source framework that orchestrates "when the AI wakes up, which channel it answers, what it stores in memory." Without this, AI only lives in one place (like a browser tab).

    View on GitHub
    📓

    Mnemon

    The memory

    Standard AI problem: every new chat starts blank — it forgets who you are. Mnemon solves this by storing facts about you in a "graph memory" — like a brain connecting related facts. Say once "I work in X" — weeks later the AI still remembers without being asked.

    View on GitHub
    👂

    Whisper.cpp

    The ears

    AI from OpenAI that turns audio into text. The .cpp version runs on small computers without sending data to external servers — so your voice notes stay private. Fits a public figure like Vivian who's sensitive about data.

    View on GitHub
    📱

    Telegram + WhatsApp (Baileys)

    The entry door

    How you actually talk to your AI — anywhere, anytime. Baileys lets WhatsApp work as a "linked device" (like WhatsApp Web). No separate bot number needed.

    View on GitHub
    📂

    Obsidian

    The filing cabinet

    Popular notes app built on Markdown files. Notes are saved as plain files, not in someone else's cloud. Works on laptop and phone, syncs automatically. AI can read + write notes to this folder.

    View on GitHub

    🐙 What you can see on Vivian's GitHub

    This is what makes it revolutionary. Vivian — a sitting minister — published all his "recipes" openly. Highly unusual.

    What he shares:

    Sit with this for a moment: a minister sharing the source code of his personal assistant. When did you last see a public official share technical knowledge like this? This is more than "open source" — it's a cultural statement: knowledge is to be shared, not monopolized.

    🚀 If you're curious and want to try (gently)

    No need to immediately build something as complex as Vivian's. Start with the smallest thing — what matters is starting.

    Level 1 — Use AI in your browser (start today)

    Open claude.ai or chat.openai.com. Start chatting. Try:

    • "Help me write an email for X"
    • "Summarize this article"
    • "I have problem Y, what's the best way to solve it?"

    This is enough for 80% of people. Vivian himself said: the tools already exist.

    Level 2 — AI on your phone via chat app

    Many people build personal Telegram bots connected to AI. The result: you can chat with AI like a friend, on Telegram. Plenty of tutorials on YouTube.

    Level 3 — Add memory

    After using AI for a while, you'll get tired of explaining who you are every new chat. This is where Mnemon or similar approaches help. Or the simplest way: make an "about-me.md" file, paste it at the start of each chat.

    Level 4 — Build a full "second brain"

    This is what Vivian did. Requires:

    • Cheap VPS (~$5/month)
    • Basic Linux & terminal knowledge
    • A few weekends
    • Willingness to read long documentation

    But don't start here. Start at Level 1. Many people stop at Level 1 and even that dramatically changes how they work.

    🎯 Why this matters — especially for us

    Vivian is actually talking about opportunity. Many countries race in the frontier of model development — building bigger, more advanced AI. That's a game of massive capital, decades of research, rare talent.

    But there's a second frontier just as important — perhaps more:

    The deployment frontier — who brings AI fastest into the daily lives of ordinary people.

    And in this race, anyone can win. Regardless of the size of your country, your office, or your savings. What decides: who dares to start trying now.

    "Don't wait for anyone. AI is in your hands today. What separates you from others isn't the tools — it's the courage to start using them."

    Want to bring AI into your business?

    30-minute free chat with the Biks.AI team. No slides, no pitch — just understanding what you need. We help Southeast Asian SMEs deploy AI agents on tools they already use.

    Book a chat