Melanchatly: home away from home

Inspiration

Moving to a new country can be one of the most life-changing and challenging experiences. Drawing from personal experiences with the emotional burden of living away from home, our team found a dire need for support in the form of an emotional outlet for international students and professionals. A recent study link indicate that as many as 92% of international students miss the comforts of home as they pursue their studies abroad; of this total, 57% miss the sensation of it, while 74% miss the sounds of their hometown. These statistics not only reflect the prevalence of homesickness but also their deep roots-sensory and emotional. The fact that this is a shared experience, added to the statistic that over 70% of international students report homesickness impacting their mental wellbeing and academic/professional performance, led us to create Melanchatly-a RAG based AI companion that understands and caters to these multi-dimensional needs of being away from home.

What it does

Melanchatly acts as an AI companion, working to understand and respond to the complex, mixed emotions associated with being away from home. Equipped with RAG-based technology, it offers empathetic conversations with native cultural context and local adaptation tips. The chatbot shares nostalgic content from users' home countries and suggests local events and communities that connect with their culture. It goes beyond mere conversation to offering practical advice on various common expatriate problems and, by the employment of RAG technology, develops personalized coping strategies. Importantly, Melanchatly is designed to identify when users may require further care, and it can help them get into contact with one of many mental health professionals or support services that might be in their area to ensure proper care is taken where needed. Thus, it is an all-inclusive support system for any homesick person abroad.

How we built it

At Melanchaty's core, we set up a RAG architecture through LangChain, exposing an optimized language model which could perform best under our resource constraints. The backend was designed in FastAPI, which provided an extremely robust and efficient way of combining our AI models with the user interface. We have carefully prepared a knowledge base about mental health resources, cultural information, and coping strategies about homesickness. This was possible through strategic prompt engineering and careful data training to enhance the model's ability for empathetic and contextually appropriate responses.

Challenges we ran into

During our development process, there were several key challenges that really put our problem-solving skills to work. Of these, the most critical was securing and curating proper training data that would facilitate substantial conversations with homesick individuals. It turned out finding quality, context-rich examples of supportive conversations was more complicated than finding reliable resources about strategies for coping with homesickness. Another important challenge was the computational limitation: we wanted to run state-of-the-art language models locally, and since computing resources were limited, this directly influenced our architectural decisions due to the need for optimization for efficiency. We also had to address cultural sensitivity and appropriateness of response against the flow of natural conversation, since homesickness can be extremely different depending on culture and other contextual factors.

Accomplishments that we're proud of

Despite the issues at the start, there were major milestones for our project that make us proud. First is the ability to implement, with success, efficient models that could serve our purpose without performance compromise. We could construct effective context layers for our chatbot through strategic system prompt engineering and making good use of available mental health data so that it could engage in substantive and empathetic conversations. This optimization not only made our solution technically viable but also saw to it that Melanchatly was in a position where it could be of emotional support and understanding to the users. The successful integration of these elements gave way to a companion chatbot balancing between technical efficiency and emotional intelligence, hence making it practical and worthy to be of use for those feeling homesick abroad.

What we learned

The development of Melanchatly proved to be an amazing learning curve that extended our horizons about the knowledge with respect to the technical and practical point of view. Indeed, this has been a deep dive into the implementation of RAG and intricacies associated with the building of empathetic AI systems. A major learning curve was how the integration at the back-end models could be done with the user interface using FastAPI; this became crucial in building a seamless experience for the users. Perhaps the most salient lesson we learned was how to truly unlock the potential of low-parameter models through efficient data training and strategic prompt engineering, allowing successful results even on limited resources. The project has also taught us the importance of cultural sensitivity in AI development and how to manage complex AI applications effectively.

What's next for melanchatly: home away from home

Looking ahead, we have plans to extend Melanchatly for even more comprehensive support. Enhancing multilingual capabilities will permit users to communicate in their native languages in a much more natural and comfortable manner. The application will be further extended by allowing voice interaction during the conversation to make it as accessible and personal as possible; therefore, users can interact with Melanchatly much more naturally. We also plan to provide a conversation history feature that will enable AI to develop deep understanding of each user's journey with them and offer more personalized support over time. We are working on the community feature that will connect users going through similar experiences, creating a supportive network for the purposes of sharing their stories and their coping ways with people who truly do understand their situation. This will further help shape Melanchatly's transformation from an AI-powered buddy to a full-fledged platform for emotional support and building communities.

Built With

• fastapi
• javascript
• langchain
• python
• rag