Our initial dataset included hotel images, which we attempted to convert into text data using an image-to-text model. However, the generated captions lacked sufficient description, so we excluded that data from our vector database.

We encountered challenges in limiting our chatbot to only address hotel and travel-related queries. Initial attempts using similarity score thresholds and varied prompting techniques to focus the bot's responses proved suboptimal. The chatbot either restricted output too much or forced irrelevant answers from our vector database. Ultimately, a simple, concise instructional approach proved most effective, allowing the bot to handle off-topic queries reasonably without overly restrictive constraints.