Garbo

inspired by

Helping smaller businesses streamline hiring candidates and organising there teams. We can reduce amount spent on human resources by executives to invest properly in the correct departments and people.

What it does

It allows users to add employees to there payroll through the chatbot. It also adds Charts allowing clean visualisation of the data stored by the software. We also have a cashflow prediction that is used on cash flow provided by the company using the product to predict future monetary power.

How we built it

Our database is built using SQLite on the backend to store employee data. This interfaces with the chatbot to allow users to use natural language to perform database operations. The chatbot can use premade tools to chain together function calls like adding 5 new people to the database all in one prompt.

We took data from the Json file storing data items (employees and there wages) and used basic charting tools to implement dynamic graphs representing data. #

The OLS regression for projection and analysis of cash flow

The development of the machine learning model had mulitple phases, experimenting with multiple technologies

• We initially started by exploring python and pytorch for machine learning model training, and discovered the library of "ONNX-GO", for cross-platform support for pre-trained neural networks.
• When considering our use of temporal data as a feature, and the ever-fluctuating nature of cash flow data in real-world examples, we decided upon Ordinary Least Squares regression. This meant a simple calculation, especially since we decided upon an analytical approach that looks at foreign trends once and determines a weight and bias from the ordinary least squares calculation once! No training loop, no evaluation. This reduced overhead and negates training time.
• This simplicity and purely mathematical approach leant us another discovery, that pytorch and onnx were not needed and OLS can be developed in native GO. Calculations including OLS and auxiliary loss functions like MSE and R^2 could all be handled with GO's math module. Said loss functions also inspired the ability to identify anomalous or even fraudulent payments.

Built With

• claude
• go
• react