Ecoceipt

Inspiration

In Indonesia, MSMEs (Micro, Small, and Medium Enterprises) make up over 99% of all businesses and employ 119 million people. Many of these businesses still manage their inventory manually using paper receipts, which are easily lost, inconsistent, and difficult to analyze. This lack of visibility often leads to overstocking, spoilage, and food waste — a growing concern both economically and environmentally. At the same time, we noticed that while AI and digital tools are widely used in large companies, there’s a huge gap when it comes to empowering small businesses with the same tools. That’s where Ecoceipt comes in — a mobile app that helps MSMEs digitize their paper receipts, understand their sales patterns, and get smart stock recommendations. The goal: reduce food waste, cut losses, and build a greener supply chain.

What it does

Ecoceipt has three core functionalities:

📸 OCR Receipt Scanner Uses Google ML Kit to capture and extract raw text from a photo of a paper receipt.
🧠 LLM-Powered Formatter Structures the unformatted text into key fields like item name, quantity, unit price, and total using the Gemini API.
📊 Stock Recommender Analyzes sales trends over time to recommend what items to restock or scale down — helping MSMEs avoid over-purchasing and wastage. All scanned and structured data is saved in Firebase for easy access, enabling businesses to keep digital records without changing their existing workflow.

How we built it

To build a solution that works for small, resource-limited businesses, we needed a tech stack that was fast to develop, simple to use, and scalable from day one. Given the tight timeframe of the hackathon, we focused on proven tools that let us move quickly without sacrificing reliability or intelligence. Our architecture balances on-device performance with cloud-based AI — so even traditionally run MSMEs can benefit from smart decision-making with just one scan.

Frontend: Kotlin with Jetpack Compose and Material3
Backend/Cloud: Firebase Firestore for database and storage
OCR: Google ML Kit Text Recognition v2
AI Formatter: Gemini API (text-to-structured output)
Architecture: MVVM with StateFlow The process starts when a user scans a receipt. The image is processed by ML Kit to extract raw text, which is then sent to the Gemini API with a custom prompt to parse it into structured fields. The result is saved to Firestore. The recommendation engine uses simple rules based on frequency and time of sales to provide weekly suggestions.

Challenges we ran into

Inconsistent receipt formats
LLM unpredictability
Camera permissions and orientation issues
Time constraints

Accomplishments that we're proud of

Built a fully working Android app in under 30 hours
Successfully integrated OCR, LLM, and Firebase in a smooth pipeline
Designed around a real-world problem with measurable sustainability impact
Created a clean, responsive UI using Jetpack Compose from scratch

What we learned

How to coordinate multiple APIs (OCR + LLM + Firebase) into a unified user flow
How to handle AI model uncertainty with human-centered fallback design
How digital tools can create real value for under-served MSMEs

What's next?

Looking ahead, Ecoceipt can expand into multilingual receipt parsing, integrate with POS systems, and provide personalized business insights over time. What started as a tool to digitize receipts can become a powerful platform for small businesses to grow sustainably — without needing to change how they work.