Akun.AI
Accounting, made reviewable
An AI-assisted accounting platform designed to make bookkeeping and financial reporting more accessible to Indonesian SMEs.
Solo product builder
Live in-development release with core accounting workflows, financial reports, an AI assistant, a receipt-OCR prototype, and Mayar subscription surfaces.
01 / 09 An AI-assisted accounting platform designed to make bookkeeping and financial reporting more accessible to Indonesian SMEs.
The problem to solve.
Akun.AI is a live, in-development AI-assisted accounting SaaS for Indonesian SMEs. It combines bookkeeping workflows, an accounting assistant, automated financial reporting, and a receipt-OCR prototype that still requires production validation.
Problem
Indonesian small and medium businesses struggle with bookkeeping complexity and lack accessible, AI-assisted accounting tools tailored to local requirements.
People
Indonesian SME owners and operators
The choices behind the interface.
Product planning, system architecture, frontend and backend development, accounting workflows, AI integration, reporting, and user experience.
- Designing accounting workflows that are accessible to non-accountants
- Implementing double-entry validation logic that catches common errors
- Building AI prompts that produce reliable, audit-ready accounting responses
What I built.
Product planning, system architecture, frontend and backend development, accounting workflows, AI integration, reporting, and user experience.
- Designed the accounting workspace architecture, including chart of accounts, ledger flows, reporting surfaces, and tenant-aware data boundaries.
- Built the full-stack SaaS flow across onboarding, dashboard, transactions, invoices, inventory, reports, AI assistant, and subscription surfaces.
- Integrated the AI assistant as a grounded finance helper that reads workspace context while deterministic validation still protects saved accounting actions.
Explore the pipeline
01Business transactions, invoices, inventory changes, and prototype receipt-OCR outputs enter the ledger workspace.
Supabase/PostgreSQL stores tenant-isolated business data; RLS boundaries protect workspace records before AI context is assembled.
02Structured transaction data is validated against chart-of-accounts rules before it reaches reports.
Structured transaction data is validated against chart-of-accounts rules before it reaches reports.
03The AI assistant answers from workspace context and summarizes financial state instead of acting as an unconstrained chatbot.
The AI assistant answers from workspace context and summarizes financial state instead of acting as an unconstrained chatbot.
04Reports, CSV exports, and dashboard cards are generated from persisted accounting records.
Double-entry checks, category validation, and permission-aware queries keep financial answers tied to stored ledger data.
Results, with context.
Live in-development release with core accounting workflows, financial reports, an AI assistant, a receipt-OCR prototype, and Mayar subscription surfaces.
Lessons & next steps +
- Accounting domain modeling requires deep understanding of Indonesian accounting standards
- AI integration for financial data needs careful grounding and validation
- Mayar billing integration requires explicit webhook and failure-path validation
Under the hood.
Technologies & features +
Next.js / TypeScript / Tailwind CSS / Supabase / PostgreSQL / Mayar / OpenRouter / Zod / Recharts
- Business onboarding and chart of accounts
- Double-entry accounting with validation
- Profit and loss reports and balance sheets
- CSV export and printable reports
- AI accounting assistant via OpenRouter
- Receipt OCR prototype for assisted transaction entry
- Team access and subscription management with Mayar
- Row-Level Security for data isolation
Feature status & limitations +
Core bookkeeping and financial reports
workingImplemented across the current pre-launch workspace; production accounting and compliance behavior still need validation.
AI accounting assistant
prototypeConnected to workspace context with deterministic validation around saved accounting actions.
Receipt OCR
prototypeThe extraction flow exists, but accuracy and failure handling are not yet production-validated.
Mayar billing
unvalidatedMayar is the configured billing provider; end-to-end payment and webhook reliability are not claimed here.
AI system notes +
Provider / model
OpenRouter-configurable LLM layer with prototype OCR ingestion for receipt capture.
Data flow
Supabase/PostgreSQL stores tenant-isolated business data; RLS boundaries protect workspace records before AI context is assembled.
Validation
Double-entry checks, category validation, and permission-aware queries keep financial answers tied to stored ledger data.
Failure handling
AI output is treated as advisory; accounting actions still pass through deterministic validation before being saved or reported.
Limitations
Pre-launch system. Tax, OCR accuracy, and local compliance behavior still need deeper production validation.
