I’m working on a solution for a client who needs an agent to pull data from a CSV file, which contains information about a provider’s location, services, categories, phone numbers, and addresses. Initial trials with chroma embeddings and a gpt-3.5-turbo LLM had inconsistent outcomes. However, switching to gpt-4-1106-preview and adjusting the chroma retriever kwargs “k” from 4 to 8 enhanced document retrieval but also increased token usage. The CSV Agent was less effective, yielding poorer results than the embeddings.
The adjunted image is a sample of my CSV/Excel file. Any advice is would be aprecciated. Excel/CSV File example
For context, my agent is an assistant that provides contact information for providers based on user queries. For example:
User: "asado barrio san vicente" AI: "Aquí tienes información sobre asado en el barrio San Vicente: Asadero Parrillero - Teléfonos: 123456789, 123456788 ASADO LA CASA DEL COSTILLAR, Javier Rodriguez - Teléfonos: 123456789 - Dirección: Example 123 Espero que esta información te sea útil. "
What should i try?