If you're inspired by our code please find a way to credit us by name :) If you'd like to hire our team or use this for research/ commercial purposes, please reach out by email.
CLICK NAME BELOW TO EMAIL: Demi Oyebanji, Daniel Froom, Anna Toaze, Suleiman Abdul Samad,
The files in this project run on the AWS System with the following programs:
- Amazon Lex
- Amazon Lambda
- Amazon S3
- Amazon SageMaker
- Amazon DynamoDB
Merlin-DRAFT is our LexBot file. In Amazon Lex, we generated a bot that speaks to users and collects info about their finances in order to assist them better. User can answer in a natural way and the bot can derive figures, approximates, labels and more and place them into a table format. The bot tries to fill out as much of this information as possible:
- Contact Information
- Country of Residence
- Location(City)
- Province/ State
- Name
- Age
- Household Size
- Housing type
- Occupation
- Income/ Income Source
- Net Liabilities
- Net Assets
- Private Medical Expenses
- Transportation Spend
- Charitable Contributions
- Essential Food Spend
- Non-essential Food Spend
- Clothing Spend
- Credit Card Debt
- Credit Score
- Debt Service
- Education Spend
- Vacation/ Travel Spend
- Wellness Spend
After the Lex bot collects all the information we need to store the information for each user in a place where we can retrieve it. Our Lambda Function connects the Lex output to DynamoDB which is an AWS-based database where we could store all user data. Once the data is accessible in Dynamo DB the magic can begin
Amazon Sagemaker allowed us to host complex Python logic on AWS which was perfect because we trained an AI to make financial predictions for a user based on lots and lots of data that we found available on Kaggle, Canadian Government Public Data, and more. It includes data like:
-
World Happiness Reports
-
Average OSAP Debt
-
USDCAD Monthly/Daily Price
-
Disribution of Household Income
-
Census Data
-
Canadian Cost Per Index Data
-
Canadaian and American Government Spend Data
-
Historical Wage Data
-
and lots, lots more
......it's a lot of CSVs User input and Canadian Data CSVs
After cleaning and preprocessing the data, we use complex machine learning models to run predictions and build clusters based on different financial patterns that overlap between datasets.
Age Profiler - 0.06 Mean Error on clustering people by age based on inputs such as income, debt, assets and more. It could form clusters such as '65+', '21-35', etc.
Income Profiler - 100% Cluster Accuracy based on inputs such as age, debt, assets, trave spending and more
In order to maintain our Notebook on Sagemaker which has all the logic, we need to keep all our CSV files on S3 so the notebook can access them.
We would have loved more time to finish the relationship between all bots but since this was a two week project we couldn't quite polish it off.
Some things we would have loved to implement:
- We expected to feed the analysis results back into Amazon Lex using another Lambda function so that the bot could give the user results
- We also wanted to extend the "conversational" feel of the bot so it feels less scripted
- The SageMaker also had an extensive timeseries logic that we would have loved to integrate with more time