This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle
What I Built
Invoice AI — Your Personal Business Analyst for Your Small Business! 🧾📊
For small shopkeepers and business owners: just click photos of your invoices on your phone, upload them, select whether each is a sales invoice or a purchase invoice, and DONE!!
Your personal business dashboard and AI-powered business analyst are ready to help you understand your business better.
Want to know which vendor accounts for most of your spending? Curious about your monthly sales trends? Have a question or just a thought in mind? Simply ask the interactive chatbot, powered by the open-weight Gemma 4 model, and get answers tailored to your business data in seconds.
But wait, I know what you might be thinking!
What about data accuracy?
After all, even a tiny digit mix-up — like 310 becoming 103 — can completely distort your business analysis. When it comes to business decisions, garbage in means garbage out!
That's why Invoice AI doesn't blindly trust extracted data. It validates invoice totals against subtotals and taxes, flags ambiguous or inconsistent extractions for review, and lets users verify uncertain information before it is confirmed and included in analytics.
The result? A smarter way for small businesses to turn everyday invoices into meaningful business insights, without manually maintaining complicated spreadsheets.
Your invoices. Your data. Your business insights. All in one place.
Demo
Video Demo:
Live Demo:
Code
💻 GitHub Repository: https://github.com/dotsource123/invoice-ai-v1.git
How I Built It
I built Invoice AI using open-source technologies, OCR, a FastAPI backend, a Streamlit dashboard, and an interactive LLM-powered assistant.
Step 1: Invoice Images → OCR
Everything starts with an invoice image captured using a phone.
- Used PaddleOCR to extract text from invoice images.
- Processed the extracted text to identify important invoice fields, including vendor details, invoice numbers, dates, subtotals, taxes, and totals.
- Converted unstructured invoice images into structured data that can be stored and analyzed.
Step 2: Validation → Reliable Business Data
Extracted numbers aren't automatically treated as correct.
- Validated invoice totals against subtotal and tax amounts.
- Flagged ambiguous or inconsistent extractions for user review.
- Kept invoices pending review out of confirmed-invoice analytics until verified.
- Distinguished purchase invoices from sales invoices so spending and sales figures are not mixed up.
- Kept different currencies separate to avoid misleading totals.
Step 3: Backend → FastAPI + SQLAlchemy
Built a FastAPI backend to handle invoice extraction, validation, storage, and dashboard analytics.
Used SQLAlchemy for database operations and created reusable business analytics functions to calculate invoice summaries, vendor spending, monthly trends, and other useful metrics.
Step 4: Interactive Dashboard → Streamlit
Built an interactive dashboard that turns confirmed invoice data into useful business information, including:
- Total purchase spending and recorded sales.
- Purchase and sales invoice counts.
- Spending by vendor.
- Monthly business trends.
- Invoice-level information for further investigation.
The goal is to make business analytics accessible even to people who don't know SQL, spreadsheets, or programming.
Step 5: AI Business Analyst → Gemma 4
This is where things get exciting!
Integrated the open-weight Gemma 4 model through Ollama Cloud to build an interactive, tool-using business assistant.
Instead of simply generating generic answers, the assistant can select from predefined business tools to retrieve relevant information from the user's saved invoice data.
For example, users can ask:
- "Which vendor accounts for most of my spending?"
- "How have my monthly purchases changed?"
- "Find invoices from a particular vendor."
- "Summarize my purchase and sales invoices."
- "Which items account for the highest recorded purchase amounts?"
The assistant uses read-only Python functions backed by SQLAlchemy queries to retrieve relevant information, then generates a natural-language response grounded in those results.
Importantly, the model does not generate and execute arbitrary SQL queries. Its access to business data is mediated through predefined tools with validated arguments.
This makes the assistant more than a chatbot: it acts as a business analyst that can retrieve relevant data and explain it conversationally.
Why Does Open Innovation Matter?
Small businesses often lack access to dedicated business analysts, expensive analytics platforms, or the technical expertise needed to make sense of their financial records.
Open innovation makes it possible to build practical AI-powered tools that bring these capabilities closer to everyday business owners.
By building around an open-weight model, open-source OCR, Python frameworks, and a tool-based architecture, I could experiment with a customizable AI assistant without making the entire application dependent on a proprietary, closed-model API.
Using Ollama Cloud also allowed me to access the model remotely rather than requiring users to run a large language model locally.
More importantly, the project demonstrates how open technologies can be combined into a complete workflow: extracting information from real-world documents, validating it, analyzing it, and making the results accessible through natural language.
My goal is simple: make useful business intelligence accessible to the smallest businesses, not just enterprises with dedicated analytics teams.
My Agent Session
[Add your DevRelay agent session link here, if available.]
Prize Categories
Best Use of Gemma
Gemma powers the AI business assistant, helping interpret business queries and generate context-aware insights from invoice data. It works alongside OCR processing and predefined business tools to turn raw invoices into actionable business intelligence, with controlled data access and validation to support security and privacy.
Best Use of Render
Deployed Invoice AI on Render, making the application accessible through the cloud without requiring users to set up a local environment. The deployment brings together the FastAPI backend and interactive dashboard, allowing users to upload invoices, explore business analytics, and interact with the AI business assistant online
