Many students have poor spending habits and losing track of one's finances may cause unnecessary stress. As university students ourselves, we're often plagued with financial struggles. As young adults down on our luck, we often look to open up a credit card or take out student loans to help support ourselves. However, we're deterred from loans because they normally involve phoning automatic call centers which are robotic and impersonal. We also don't know why or what to do when we've been rejected from loans. Many of us weren't taught how to plan our finances properly and we frequently find it difficult to keep track of our spending habits. To address this problem troubling our generation, we decided to create AvaAssist! The goal of the app is to provide a welcoming place where you can seek financial advice and plan for your future.


AvaAssist is a financial advisor built to support young adults and students. Ava can provide loan evaluation, financial planning, and monthly spending breakdowns. If you ever need banking advice, Ava's got your back!


🧠UX Research🧠

To discover the pain points of existing banking practices, we interviewed 2 and surveyed 7 participants on their current customer experience and behaviors. The results guided us in defining a major problem area and the insights collected contributed to discovering our final solution.

💸Loan Research💸

To properly predict whether a loan would be approved or not, we researched what goes into the loan approval process. The resulting research guided us towards ensuring that each loan was profitable and didn't take on too much risk for the bank.


✏️UI/UX Design✏️

alt text

Figma was used to create a design prototype. The prototype was designed in accordance with Voice UI (VUI) design principles & Material design as a base. This expedited us to the next stage of development because the programmers had visual guidance in developing the app. With the use of Dasha.AI, we were able to create an intuitive user experience in supporting customers through natural dialog via the chatbot, and a friendly interface with the use of an AR avatar.

Check out our figma here

Check out our presentation here

📈Predictive Modeling📈

The final iteration of each model has a test prediction accuracy of +85%!

alt text

We only got to this point because of our due diligence, preprocessing, and feature engineering. After coming up with our project, we began thinking about and researching HOW banks evaluate loans. Loan evaluation at banks is extremely complex and we tried to capture some aspects of it in our model. We came up with one major aspect to focus on during preprocessing and while searching for our datasets, profitability. There would be no point for banks to take on a loan if it weren't profitable. We found a couple of databases with credit card and loan data on Kaggle. The datasets were smaller than desired. We had to be very careful during preprocessing when deciding what data to remove and how to fill NULL values to preserve as much data as possible. Feature engineering was certainly the most painstaking part of building the prediction model. One of the most important features we added was the Risk Free Rate (CORRA). The Risk Free Rate is the rate of return of an investment with no risk of loss. It helped with the engineering process of another feature, min_loan, which is the minimum amount of money that the bank can make with no risk of loss. Min_loan would ultimately help our model understand which loans are profitable and which aren't. As a result, the model learned to decline unprofitable loans.

alt text

We also did market research on the average interest rate of specific types of loans to make assumptions about certain features to supplement our lack of data. For example, we used the average credit card loan interest rate of 22%. The culmination of newly engineered features and the already existing data resulted in our complex, high accuracy models. We have a model for Conventional Loans, Credit Card Loans, and Student Loans. The model we used was RandomForests from sklearn because of its wide variety of hyperparameters and robustness. It was fine-tuned using gridsearchCV to find its best hyperparameters. We designed a pipeline for each model using Pipeline, OneHotEncoder, StandardScaler, FunctionTransformer, GradientBoostingClassifier, and RandomForestClassifier from sklearn. Finally, the models were saved as pickle files for front-end deployment.

🚀Frontend Deployment🚀

Working on the frontend was a very big challenge. Since we didn't have a dedicated or experienced frontend developer, there was a lot of work and learning to be done. Additionally, a lot of ideas had to be cut from our final product as well. First, we had to design the frontend with React Native, using our UI/UX Designer's layout. For this we decided to use Figma, and we were able to dynamically update our design to keep up with any changes that were made. Next, we decided to tackle hooking up the machine learning models to React with Flask. Having Typescript communicate with Python was difficult. Thanks to these libraries and a lot of work, we were able to route requests from the frontend to the backend, and vice versa. This way, we could send the values that our user inputs on the frontend to be processed by the ML models, and have them give an accurate result. Finally, we took on the challenge of learning how to use Dasha.AI and integrating it with the frontend. Learning how to use DashaScript (Dasha.AI's custom programming language) took time, but eventually, we started getting the hang of it, and everything was looking good!


  • Our teammate, Abdullah, who is no longer on our team, had family issues come up and was no longer able to attend HackWestern unfortunately. This forced us to get creative when deciding a plan of action to execute our ambitious project. We needed to redistribute roles, change schedules, look for a new teammate, but most importantly, learn EVEN MORE NEW SKILLS and adapt our project to our changing team. As a team, we had to go through our ideation phase again to decide what would and wouldn't be viable for our project. We ultimately decided to not use Dialogflow for our project. However, this was a blessing in disguise because it allowed us to hone in on other aspects of our project such as finding good data to enhance user experience and designing a user interface for our target market.
  • The programmers had to learn DashaScript on the fly which was a challenge as we normally code with OOP’s. But, with help from mentors and workshops, we were able to understand the language and implement it into our project
  • Combining the frontend and backend processes proved to be very troublesome because the chatbot needed to get user data and relay it to the model. We eventually used react-native to store the inputs across instances/files.
  • The entire team has very little experience and understanding of the finance world, it was both difficult and fun to research different financial models that banks use to evaluate loans.
  • We had initial problems designing a UI centered around a chatbot/machine learning model because we couldn't figure out a user flow that incorporated all of our desired UX aspects.
  • Finding good data to train the prediction models off of was very tedious, even though there are some Kaggle datasets there were few to none that were large enough for our purposes. The majority of the datasets were missing information and good datasets were hidden behind paywalls. It was for this reason that couldn't make a predictive model for mortgages. To overcome this, I had to combine datasets/feature engineer to get a useable dataset.


  • Our time management was impeccable, we are all very proud of ourselves since we were able to build an entire app with a chat bot and prediction system within 36 hours
  • Organization within the team was perfect, we were all able to contribute and help each other when needed; ex. the UX/UI design in figma paved the way for our front end developer
  • Super proud of how we were able to overcome missing a teammate and build an amazing project!
  • We are happy to empower people during their financial journey and provide them with a welcoming source to gain new financial skills and knowledge
  • Learning and implementing DashaAi was a BLAST and we're proud that we could learn this new and very useful technology. We couldn't have done it without mentor help, 📣shout out to Arthur and Sreekaran📣 for providing us with such great support.
  • This was a SUPER amazing project! We're all proud to have done it in such a short period of time, everyone is new to the hackathon scene and are still eager to learn new technologies


  • DashaAi is a brand new technology we learned from the DashaAi workshop. We wanted to try and implement it in our project. We needed a handful of mentor sessions to figure out how to respond to inputs properly, but we're happy we learned it!
  • React-native is a framework our team utilized to its fullest, but it had its learning curve. We learned how to make asynchronous calls to integrate our backend with our frontend.
  • Understanding how to take the work of the UX/UI designer and apply it dynamically was important because of the numerous design changes we had throughout the weekend.
  • How to use REST APIs to predict an output with flask using the models we designed was an amazing skill that we learned
  • We were super happy that we took the time to learn Expo-cli because of how efficient it is, we could check how our mobile app would look on our phones immediately.
  • First time using AR models in Animaze, it took some time to understand, but it ultimately proved to be a great tool!

⏭️WHAT'S NEXT FOR AvaAssist⏭️

AvaAssist has a lot to do before it can be deployed as a genuine app. It will only be successful if the customer is satisfied and benefits from using it, otherwise, it will be a failure. Our next steps are to implement more features for the user experience.

For starters, we want to implement Dialogflow back into our idea. Dialogflow would be able to understand the intent behind conversations and the messages it exchanges with the user. The long-term prospect of this would be that we could implement more functions for Ava. In the future Ava could be making investments for the user, moving money between personal bank accounts, setting up automatic/making payments, and much more.

Finally, we also hope to create more tabs within the AvaAssist app where the user can see their bank account history and its breakdown, user spending over time, and a financial planner where users can set intervals to put aside/invest their money.


Yifan is a 3rd year interactive design student at Sheridan College, currently interning at SAP. With experience in designing for social startups and B2B software, she is interested in expanding her repertoire in designing for emerging technologies and healthcare. You can connect with her at her LinkedIn or view her Portfolio

Alan is a 2nd year computer science student at the University of Calgary. He's has a wide variety of technical skills in frontend and backend development! Moreover, he has a strong passion for both data science and app development. You can reach out to him at his LinkedIn

Matthew is a 2nd year student at Simon Fraser University studying computer science. He has formal training in data science. He's interested in learning new and honing his current frontend skills/technologies. Moreover, he has a deep understanding of machine learning, AI and neural networks. He's always willing to have a chat about games, school, data science and more! You can reach out to him at his LinkedIn

📣📣 SHOUT OUT TO ABDULLAH FOR HELPING US THROUGH IDEATION📣📣 You can still connect with Abdullah at his LinkedIn He's super passionate about reactJS and wants to learn more about machine learning and AI!


Built With

Share this project: