Botzy.ai is the most significant project I have worked on because it brought together many of the concepts I am most interested in: AI systems, retrieval-augmented generation (RAG), software engineering, and real-world reliability. The goal of Botzy is to help small businesses deploy AI chatbots that answer customer questions using only approved business documents. While building it, I learned that creating an AI application is about much more than choosing a model—it requires careful attention to data retrieval, system design, evaluation, and user trust.

One of the most valuable lessons came from analyzing a failure case within the system. In my example, a customer asks whether a personalized item can be returned. The chatbot retrieves the general return policy stating that returns are accepted within 30 days, but it fails to retrieve an important exception stating that personalized items are final sale. Because the retrieval step provided incomplete context, the model generated a confident but incorrect answer.

This analysis showed me that AI failures often originate from system design rather than model capability. Even when a response is grounded in retrieved documents, missing critical information can lead to operationally incorrect outcomes. The experience reinforced the importance of evaluation, human oversight, and robust retrieval pipelines when building trustworthy AI systems.