The Challenge

Most Indonesians who want to budget rely on manual spreadsheets or note apps, and the effort of logging every transaction is exactly why the habit collapses within weeks. Expense tracking competes with real life: people spend many small amounts every single day (a coffee here, an ojek fare there), and each one requires opening an app, choosing a category, and typing an amount. That repeated friction, multiplied across dozens of transactions per week, means data goes stale, categories get inconsistent, and users lose any real visibility into where their money goes. Without a frictionless capture method, personal budgeting stays a good intention rather than a habit.

Constraints & Scale

FinCuy had to work for non-technical users on the tools they already own, which ruled out any friction-heavy onboarding or a separate mobile app to install and maintain. Telegram was chosen as the single interface, so every interaction had to feel instant and forgiving of messy input like typos, slang, and shorthand amounts (25k, 25rb, 25.000). As a solo-built product, the stack had to stay lean enough for one developer to operate and extend, while still managing AI inference and OCR costs per message in a way that keeps a Rp 20,000 monthly plan sustainable. Billing, trials, and per-tier limits all had to be enforced reliably from day one.

System Architecture

FinCuy runs as two coordinated services over a shared PostgreSQL database. A Node.js bot service handles the Telegram webhook, receives text and image messages, and orchestrates the extraction pipeline. Free-text messages are parsed by a large language model that infers the amount, category, and date, while receipt photos are routed through OCR before the same normalization layer maps them onto a fixed 8-category schema. Structured transactions are then persisted to PostgreSQL, which is the single source of truth for expenses, categories, recurring costs, subscriptions, and trial or billing state. A separate Next.js and TypeScript web dashboard reads from the same database to render real-time charts, category breakdowns, and monthly summaries, styled with Tailwind CSS. Because both the bot and the dashboard share one schema, data stays consistent regardless of which surface created or viewed it. Subscription tiers (free trial, Basic, and full package) are enforced in the bot layer so that feature access and usage limits are checked before any AI or OCR call is made.

Key Tradeoffs

We chose Telegram as the only interface over building a native mobile app because it removed install friction entirely and let a solo team ship and iterate without app-store review cycles. We chose an LLM plus OCR extraction pipeline over rigid keyword rules because real user input is messy (slang, typos, mixed formats like 25k and 25rb), and a rule-based parser would break on the long tail of how people actually write. We chose to normalize every AI output against a fixed category schema over trusting raw model output because analytics are only useful if categories stay consistent, and a slightly constrained schema produces far more trustworthy dashboards than open-ended labels. The honest trade-off: per-message AI and OCR calls add real cost, so tier limits exist specifically to keep the economics viable at a Rp 20,000 price point.

Measurable Results

FinCuy compresses expense logging from a multi-step spreadsheet routine to a single Telegram message completed in under 10 seconds. Onboarding friction drops to effectively zero: there is no app to install and no signup form, so users log their first expense from message one. The AI pipeline auto-classifies every entry across 8 categories, removing manual tagging entirely, and two capture modes (text and receipt photo) cover the full range of daily spending. Pricing stays accessible at Rp 20,000 per month after a 7-day free trial, making structured personal finance tracking realistic for everyday users rather than a chore they abandon.

Case Study

FinCuy: AI-Powered Personal Finance Bot

2026

FinCuy logs and categorizes expenses from a single Telegram message or receipt photo, cutting manual bookkeeping to under 10 seconds per entry.

FinCuy: AI-Powered Personal Finance Bot screenshot 1

FinCuy: AI-Powered Personal Finance Bot screenshot 2

FinCuy: AI-Powered Personal Finance Bot screenshot 3

FinCuy: AI-Powered Personal Finance Bot screenshot 4

01/04

Tech Stack

Telegram Bot API

AI / LLM + OCR

Key Features

01Logs expenses instantly from a plain Telegram chat message
02Scans photographed receipts with OCR and auto-fills the details
03Categorizes every transaction automatically across 8 spending categories
04Detects dates and recurring costs without extra user input
05Visualizes spending in a real-time web dashboard with charts
06Understands messy input like slang, typos, and shorthand amounts
07Syncs every entry across devices with no app install required

Project Details

FinCuy is an AI-powered personal finance assistant that converts a single Telegram message or a photographed receipt into a fully categorized expense record in under 10 seconds. It is built for Indonesian users who find spreadsheet budgeting tedious and inconsistent, and it removes the friction of manual tracking by living inside a messaging app people already open dozens of times a day. There is nothing to download and no account setup ritual: users send their first expense and start tracking immediately. Aksara Karya designed and built FinCuy end to end as a solo full-stack product, spanning the Telegram bot, the AI processing pipeline, the web dashboard, and subscription billing. The product ships with 6 core capabilities and classifies spending across 8 categories automatically. Two capture modes cover the way people actually spend: natural-language chat such as "kopi 25rb" for quick entries, and photo capture for physical receipts. Smart date handling and recurring-cost detection mean users do not have to restate context every time, which keeps daily logging effortless. The architecture separates a Node.js Telegram bot service from a Next.js and TypeScript dashboard, with PostgreSQL as the single source of truth for transactions, categories, subscriptions, and billing state. The central engineering challenge was reliable extraction: turning unstructured free text and photographed receipts into clean amount, category, and date fields. FinCuy pairs a large language model for intent and category inference with OCR for receipt images, then validates every result against a fixed category schema before it reaches the database. This normalization step is what keeps analytics trustworthy, because a dashboard is only as good as the consistency of the data behind it. The outcome is a workflow that compresses expense logging from a multi-step spreadsheet routine down to one message, typically completed in under 10 seconds. Pricing starts at Rp 20,000 per month after a 7-day free trial, with a Rp 35,000 tier that unlocks advanced multi-receipt OCR and custom categories, keeping the product affordable for individual users. Every entry syncs to a real-time dashboard where category breakdowns and monthly trends are visible from any device, turning scattered daily spending into a clear financial picture.

#fintech#ai#telegram-bot#nextjs#ocr#saas

Try FinCuy free for 7 days at fincuy.my.id, or ask Aksara Karya to build your AI product.

Next Step

Start Your Digital Transformation

Professional websites & apps to boost leads and sales. We handle everything.

Consult Now

More Projects6 projects

View Case Study

KRW Renovasi Contractor Marketing Website

2026

A local-SEO marketing website for KRW Renovasi, a Karawang contractor, with 6 service pages and a WhatsApp-first lead flow built on Next.js.

Live Demo View Case Study

View Case Study

VPS Elektronika – EV Bus Distributor Website

2026

Professional WordPress website for PT VPS Elektronika Indonesia, showcasing 2 EV bus product lines and 6 services, delivered in under 3 weeks.

Live Demo View Case Study

View Case Study

SESPIM LEMDIKLAT POLRI – WordPress Security Recovery

2026

Remediated a 3-year silent WordPress breach for Indonesia's police training institution, removing 13 malware files and restoring full access in 24 hours.

Live Demo View Case Study