Vita | Devpost

Portada del proyecto

Inspiration

Onboarding processes are crucial for attracting new clients and retaining existing ones; however, they often face several challenges, such as a lack of automation, security issues, and a poor user experience. To overcome these obstacles, it is important that companies focus on integrating advanced technologies like AI and machine learning to automate tasks and ensure a secure onboarding journey for all types of users. Additionally, they need to provide personalized support and clear communication throughout the process to enhance user satisfaction and engagement. This will lead to a higher retention and better customer relationships. We also heavily consider that nowadays, it is very complicated for young people to make right decisions about their finances, such as which debit or credit card they should get, which investment makes more sense for them, and the many different financial terms that are not easily comprehended by the average person. These two problems is what Vita tackles, meaning lighthouse in norwegian!!

What it does

We created an onboarding experience that focuses on creating the ideal experience for costumers depending on their demographic sector and different factors such as age, entity they live in and income. This results in a higher user satisfaction in which the client will receive the best option Banorte can give him based on the factors mentioned previously.

How we built it

Using different technologies such as AI intelligence, data analytics, face recognition and image scanning, we were able to create this onboarding process that uses various libraries like: Tesseract OCR, Open AI, pandas, deepface and keras Tensorflow.

Challenges we faced

There were many problems in the process of developing Vita. For example, at the beginning we thought MongoDB would make for a great database to store client data, however, after many hours of trying to get it to work, we realized that MongoDB was very hard to use alongside Django, and so we had to drop the idea and use SQLite.
Another important challenge was getting the data on an image to text. For this we used Tesseract, which does exactly that, however, due to the fact that the composition of the INE is very irregular, the alorithm quickly got confused and was not able to return the necessary text. To fix this, we cropped the image using proportions, so that less text would contribute to the overall noise of the image. On top of this, we used openai chat-gpt 3.5 to clean the data and rewrite it using logic, an example of this would be that in the CURP, zeros and o's would get confused pretty easily, but we knew the first 4 characters could only be a letter, so it had to be an o and so on.
Working the front end without a framework was also pretty complicated, since we ended up with a lot of htlm files to fit the tailored home pages. This ended up being very time consuming however, in the end, we became quite efficient doing it. The reason for not using a framework was the fact that in the past, we used Streamlit for the front end, which only needs basic python knowledge to work, however, for this event, we wanted to strive further and built our complete front end using html and css.
Finally, another big problem was the fact that since we did not have a lot of mastery in the realm of front-end, it became a bit complicated to feed the data from the back-end into the front-end, this ended up in a lot of researching and a lot of trial and error until we ended up with very nice understanding on how to develope a webpage.

Our achievements

Like we mentioned in the last paragraph, we entered the event with no previous experience developing the front-end in a hackathon, and even though it arised the majority of our problems, it was also one of the most rewarding parts of the whole event, since we had the opportunity to learn it from top to bottom.
We also used new libraries for the first time, some of them even being a core part of our back-end, such as tesseract or open-ai. This was also very rewarding since we had to adapt the strategy using those libraries in creative ways rather than ditching the idea and trying a simpler or a most familiar way.
Additionally, this challenge was full of AI and ML, which is something we are really interested about, using algorithms like Keras to train a virtual assistant or understanding the logic behind open-ai queries was truely something mind-breaking.
Finally, this challenge was not only about programming, it was also about learning something brand new for the most of us: finance! Having the challenge of learning a whole new discipline and then wrapping your mind around it to end up with complicated algorithms, flow diagrams and ideas to compliment the theory you have just aquired was quite complicated but something refreshing and satisfying.

Next steps

We would love to have more time to keep on developing, as well as deep-diving into the financial part of our project, since our understanding on finance was very limited at the beggining of the event, we would like to take a deeper dive into investments and how they relate to risk in more elements apart from age and income. Talking about tech, we would like to improve the design of the page using a framework like React to make it simpler and cleaner. The same goes for the back-end, which insted of using approximations for which sections of the ids to crop, use smart references such as marks on the ID to know where to crop for maximum accuracy. Lastly, a lot of work has to be put into the back-front-end-bridge (quite a word, right?), mainly we would like to add the ability to scan the ID rather than uploading a photo, skipping some tedious processed for the client and slowing the overall flow of the onboarding.