This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend

What I Built

My father runs a small local bakery and snack shop, and for years his accounts have lived in a notebook. Every day he writes down what he bought for the shop (snacks, milk, drinks, samosa and more), how much cash he received, how much came in online, and how much he set aside for tomorrow. At the end of the month, he adds it all up by hand.

I built Bakery Daily Helper to make that routine easier. He types the day's note the way he already writes it, for example:

snacks 3456, milk 800, samosa 500, cash 6100, online 2000, kept 2000

The app reads the note, fills in a table, and lets him check and correct it before saving. From there, he gets:

• A monthly report with total credited (cash + online), total spent on stock, and the difference
• Charts showing spending by category and money in and out, day by day
• Ask your notebook, where he can ask questions like "How much did I spend on milk last week?"
• What to load next, with simple suggestions based on the last seven days

Daily Book:

Month report:

Ask your notebook:

What to load next:

All screenshots and the video use fake sample data, not real numbers from the business.

Demo

Code

How I Built It

The stack is simple:

• Gemma, Google's open-weight model, running locally through Ollama
• Streamlit for the interface and pandas for the calculations
• Two small CSV files on the laptop to store the data

Gemma has two jobs in this project. First, it reads the daily note and turns it into a structured table. Second, it turns a question such as "Which category did I spend the most on this month?" into a small query. Python then runs that query on the saved data and finds the exact number, and Gemma explains the result in a friendly sentence.

I let Python handle all the maths on purpose. Small models are not reliable with arithmetic, and a shopkeeper's totals have to be exactly right.

A few things I learned along the way:

• Model size matters. With the smallest model (Gemma 1B), the app sometimes skipped items in a note and filed a whole amount under "Other spendings". Switching to Gemma 4B on my 8 GB laptop and tightening the prompt fixed this. The app also shows a table before saving, so any mistake can be corrected by hand.
• I had the data model wrong at first. I assumed cash and online money were recorded for each category. My father pointed out that they are whole-day totals, and only purchases are per category. I rebuilt the app around that, and it became far more realistic.
• I used an AI assistant to help write and debug parts of the code, then tested and adjusted everything myself.

Why Does Open Innovation Matter?

My father's notebook is his business. It shows what he buys, what he earns, and how his shop is doing. With a closed AI service, all of that would travel to someone else's server, and it would need an internet connection and probably a subscription. With Gemma running on the laptop, nothing leaves the machine, it costs nothing to run, and it works with no internet. For a small shop, that matters.

Being open also gave me control. I could swap Gemma 1B for 4B, rewrite the prompt, and match the categories to his real shop without asking anyone for permission.

To be honest, a closed model would probably be smarter and misread fewer notes. Here, privacy, zero cost and offline use were worth more than raw intelligence. And because Python does the calculations, the totals are always exact, and the table he checks before saving lets him catch any note the model misreads.

What My Father Said

I showed him the app, and he liked it. He told me it is now much easier to understand how the shop is doing through the charts and graphs: he can quickly see what is going up or down, from his spending by category to the money coming in each day, and he can check the monthly credited money without any effort. Before, he had to keep records manually in a notebook. Now every entry is saved automatically. He also liked that the app can show patterns in the middle of the month, such as where to look more closely and where to invest.

He gave me honest feedback as well. Some of the tips on the "What to load next" page weren't realistic for a shop like his. They sounded like financial plans that, from his experience, would never work in practice. He still found the app useful, but his comment taught me something important: AI suggestions are only a starting point, and a shopkeeper's experience matters more. Making those suggestions more realistic is the first thing I want to improve.

What's Next

• Reading a photo of the notebook page instead of typing
• Suggestions that reflect real shop conditions, using my father's own corrections
• Voice entry, so he can simply speak the day's note

Prize Categories

• Gemma (partner technology)

Thank you for reading, and thanks to the Hacktoberfest team for the live streams that helped me get started.