Virgil

Inspiration

We were motivated by the mounting issue of medical fraud that impacts billions of dollars every year and disproportionately impacts seniors. We wanted to design something user-friendly but efficient enough to let people authenticate prescriptions and medical services in real-time and stifle fraud in their tracks. And in the hopes to adequately benefit the community in a way which represented our beliefs and skillset, we created Virgil

What it does

Virgil scans medical records, prescriptions, and bills to check if they are valid and to prevent healthcare fraud against users. It also features a chatbot using artificial intelligence to make users aware of scams and a reporting function to report fraud if suspected.

How we built it

We built Virgil with React Native Expo with React Native Paper styling to make it cross compatible with various platforms and included a document scanner and machine learning chatbot. We are using flask for our backend withwith SQLite to store relevant data, Redux for caching, and all of it is integrated with medical databases to authenticate. The ml models include a modified oLlama with RAG + Langchain for document context, and OpenAI for the general model. Data vectorization and vector embeddings were utilized for optimal communications with models.

Challenges we ran into

Among the essential challenges was to incorporate a quality document scanner. We tried several attempts with other solutions, yet in the end settled on switching from Springboot to Flask for easier document parsing, and then vectorization of data and processing done via llm vector embeddings for optimal communication with the llms. We also bridged UI/UX issues to ensure user-friendliness, particularly to senior users. The fact that Langchain is relatively new and lacked a mature ecosystem was our main issue, however we were able to counteract it via using courses and university lectures on the subject.

Accomplishments that we're proud of

We were able to create an easy-to-use tool that can prevent medical fraud through real-time verification. Our incorporation of AI and fraud reporting brings it to the next level by not making it just a detection tool but a preventive solution. And overall making a potent app that benefits the community, represents our skillset, and works with state of the art technology. And, our most impressive achievement being that we got a NCDHHS Senior interested enough in our project to do an interview.

What we learned

We discovered backend databases like SQLite and became more proficient at dividing work efficiently. We found good user design and ways in which real-world input can drive product design. We also learnt when to utilize various backends for their strengths, and how python w/ flask is optimal for data intensive projects.

What's next for Virgil

We will incorporate voice navigation into the app to make it more accessible, forge more hospital partnerships and with medical agencies, and have a more robust fraud-reporting mechanism so that law enforcement can prosecute medical fraud more effectively. As well as to improve on UI and potentially use more potent models such as deepseek r1, grok 3, or OpenAI's o lineup of models.