Rangka AI designs a DIY blueprint for your own use case: what to build, a simple diagram of the tools, ready-to-paste prompts for your own Claude or ChatGPT, and a testing and safety checklist with PDPA and API-key checks built in. Every custom DIY blueprint is test-built from the prompts alone in a test environment with simulated services and fake data, before release. Ready-made products come with ready code instead of build prompts and are checked separately; each product shows its own test status.
For individuals, freelancers and small businesses in Malaysia. Every engagement starts with a short intake form, no sales call needed. Email us and we'll send it to you.
“Rangka dulu, baru bina.” Plan the framework first, then build.
Maybe you already use ChatGPT for the odd email. Maybe you've tried asking it to “build me a bot” and got pages of code you can't check. There are questions nobody has answered properly:
You're not alone. In a survey of 133 Malaysian SMEs:1
Most automation providers in Malaysia are build shops, and scoping is often bundled into the sale. We design first, and building with us is never required.
Rangka means framework. Every DIY blueprint has four parts, written in plain language for the person who will build it.
The problem and the solution, including what the system must never do.
A simple diagram and the tools (for example Google Sheets, Telegram, Apps Script), explained without jargon.
Step-by-step prompts you paste into your own Claude or ChatGPT, each followed by what to check.
Functional and prompt-injection tests, API-key and access checks, and PDPA items.
Tested before release. Every custom DIY blueprint is test-built from the prompts alone in a test environment with simulated services and fake data, before release. Ready-made products come with ready code instead of build prompts and are checked separately; each product shows its own test status. If our tester can't build a custom DIY blueprint from its prompts, the prompts aren't ready, and we fix them before you get them. This does not test your live accounts, AI model version or language understanding.
The lowest-cost way in: a free ready-made guide for one common use case, written once for many people, with an optional code bundle. When your process is different, step up to a Quick-Scan or a Custom DIY Blueprint below.
Code bundle
RM150
Code.gs, licence, README and PDF guide, sold on Gumroad
Buy the code bundle (RM150)Status: on sale. Checked with a simulated test harness only; not yet run on live services.
More ready-made blueprints, and an e-book, to follow.
“Where do I start?”
“We know what to automate; we want to build it ourselves.”
For firms with 50–200 staff already using AI informally.
Build review: we check what you or your builder built against the blueprint's checklist before go-live. RM150–300/hour.
Monthly advisory retainer: RM800–2,000/month.
Optional build service: optional, on request, quoted separately, needs a separate agreement. Never bundled into the blueprint; its scope (including whether any hosting or ongoing support is included) is set out in that separate agreement.
Conflict of interest. We also offer an optional build service. To keep our advice independent: the blueprint is written so anyone can build it; you're free to use any builder; we never recommend a build from us in the blueprint; and the build quote is separate and optional.
Prices are indicative ranges. We confirm a fixed written quote after the intake form and clarifying round, before payment and before any work starts.
You build it with the AI tools you already pay for, step by step, so you understand what you built and can fix it later.
Every custom DIY blueprint includes a PDPA and AI-risk section that covers wrong answers (hallucination), prompt injection, bias, API keys and access, and where a human must stay in the loop. Under the amended PDPA, penalties go up to RM1 million or 3 years' imprisonment, and notifiable data breaches must be reported to the Commissioner within 72 hours.2 It is much easier to plan for that before a build than after.
Every custom DIY blueprint is test-built from the prompts alone in a test environment with simulated services and fake data, before release. We fix the prompts until it works. Ready-made products come with ready code instead of build prompts and are checked separately; each product shows its own test status.
We don't resell tools, and we accept no referral fees or commissions. Advisory by default; building is an optional, separate service, and we never recommend a build from us in the blueprint (see the disclosure above).
Our ROI estimates show the downside as well as the upside. If an automation does not clearly pay for itself, we say so.
We work from process descriptions and anonymised samples under an NDA/data-processing agreement, and delete working files after the engagement. Our build prompts never ask you to paste real customer data, tokens or keys into an AI chat.
Anonymised and based on a real system; no real business, people or data. Technicians report expense claims, job records and parts purchases by chatting to a Telegram bot in Malay, English or a mix, including receipt photos and voice notes. Nothing is saved until they tap Confirm.
Excerpt from the ground-rules prompt:
Never put the Telegram bot token in any URL or text that is sent to the AI provider. To give an image or audio file to the AI, download it inside Apps Script and send the bytes (base64 / file upload), never the Telegram file link.
Status: Passed our build test: test-built from the prompts alone with simulated services and fake data; not yet run on live services.
Automated step Human decision
Client document intake, chasing & e-Invoice checks for a fictional KL accounting firm: three cooperating agents that file documents, chase what's missing and flag e-Invoice mismatches, while accountants approve every entry and nothing is submitted to LHDN by the AI. This earlier-format sample shows the design, PDPA and risk sections, without build prompts.
Download sample (PDF)Samples are clearly marked as fictional or anonymised. All volumes and costs in them are illustrative.
Email us and we'll send it. Tell us how the work runs today. Process descriptions are enough; we don't need your customer data.
Our questions, answered by email. For Blueprint and Roadmap you can also ask for one call of 30 minutes max. The Quick-Scan needs no call.
Within the ranges above.
Work starts once payment is received.
Your report, or your blueprint test-built from the prompts alone in a test environment with simulated services and fake data, before we deliver it. Then you build it with your own Claude or ChatGPT and work through the checklist. Want a second pair of eyes? Add a build review. Want us to build it? Ask about the optional build service (optional, on request, quoted separately, needs a separate agreement).
Who want to build AI tools for their own work, properly.
Clinics, accounting and professional firms, e-commerce sellers, F&B outlets and education or tuition centres, especially where the work runs on WhatsApp, Telegram, email and spreadsheets.
Whose teams already use AI informally and now need a policy, a risk register and a roadmap.
Probably not for you if: you want a system delivered and run for you with no involvement. Our build service is optional, on request, quoted separately and needs a separate agreement; our main product is a blueprint you build yourself.
No coding, but some setup (accounts, copy-paste, following steps). If that's not for you, ask for a build review or about the optional build service.
Your own Claude or ChatGPT account. The system you build may also need a pay-as-you-go AI provider account; the blueprint tells you how to set a spending limit.
Not by default. Advisory by default; building is an optional, separate service: optional, on request, quoted separately, needs a separate agreement. We can also review what you built before go-live as an add-on.
We also offer an optional build service. To keep our advice independent: the blueprint is written so anyone can build it; you're free to use any builder; we never recommend a build from us in the blueprint; and the build quote is separate and optional.
Every custom DIY blueprint is test-built from the prompts alone in a test environment with simulated services and fake data, before release. Ready-made products come with ready code instead of build prompts and are checked separately; each product shows its own test status. If a custom DIY blueprint can't be built from its prompts, we fix the prompts first. This does not test your live accounts, AI model version or language understanding. It's a quality check on the blueprint, not a guarantee that your build will behave identically with your own accounts and data.
Whatever fits. We aren't tied to any vendor, and we prefer tools you already have.
That's not how we design. Every custom DIY blueprint keeps a human in the loop for decisions, approvals and anything sensitive.
Every custom DIY blueprint includes a PDPA and AI-risk checklist; each ready-made product includes its own PDPA and data safety page. During the engagement we ask for process descriptions and anonymised samples rather than real data, work under an NDA/data-processing agreement, and delete working files afterwards.
Probably not for the advisory work itself, so please check with the agency first. A build by an eligible provider may qualify for schemes such as the MSME Digital Grant MADANI (50% of eligible invoices, up to RM5,000, subject to eligibility).3
Quick-Scan report: within 3 working days of payment. Custom DIY Blueprint: 2–3 weeks from payment. AI Roadmap + Governance: 6–8 weeks from payment. Before that: the intake form, one written clarifying round and a fixed written quote.
You do, once it's paid for.
No. We design processes and automations. Professional decisions stay with you and your licensed advisers.
Yes. Intake forms, written questions, optional calls, and each blueprint's explanations and checklist are available in English or Bahasa Melayu; the build prompts stay in English (the tested version) unless a Bahasa Melayu prompt set has passed its own test build.
Boleh. Borang maklumat awal, soalan bertulis, panggilan pilihan (jika diminta), serta penerangan dan senarai semak dalam setiap blueprint boleh disediakan dalam Bahasa Inggeris atau Bahasa Melayu; prompt binaan kekal dalam Bahasa Inggeris (versi yang telah diuji) kecuali set prompt Bahasa Melayu telah lulus ujian binaannya sendiri.
Start with an AI Quick-Scan. Within 3 working days of payment you'll have 3–5 prioritised use cases, 2–3 quick wins you can try this week, and the risks to watch for.
Or write to hello@rangkaai.my