I am developing a chatbot using ChatGPT, which recommends recipes to users. A unique challenge I face is ensuring ChatGPT suggests recipes based solely on the ingredients we currently have in stock. Since our inventory changes daily, ChatGPT needs to be updated with our inventory data once a day to provide accurate recommendations. Thus, if a user asks for Indian recipes, ChatGPT should be aware of and suggest recipes that align with that day’s available ingredients.
Initially, I attempted integrating Azure Cognitive Search with ChatGPT. When querying about specific products, the system seemed to function adequately. However, when users ask about recipes, the response is often inaccurate or suggests ChatGPT is unaware of the correct answer.
Subsequently, I tried using ChatGPT with the Langchain method to process user input into queries. This used many tokens and often produced inaccurate queries that didn’t match what we expected.