I'm developing a SaaS chatbot service that aims to provide individualized ChatGPT models for hundreds of customer accounts. Each customer should be able to train their ChatGPT model with their own knowledge base, including documents in various formats (e.g., doc, pdf, txt) to cover specific information like price lists or FAQs. I'm exploring the best method to achieve this at scale, considering three main approaches:
Entitlements: I'm unsure if this would effectively manage access at the scale and customization needed.
Prompt Engineering: By integrating the knowledge base summary into the prompt, but concerned about information loss.
Fine-Tuning: Uncertain about the feasibility of fine-tuning multiple GPT-3.5 models for each account and then utilizing them effectively.
A similar feature is offered by services like CustomGPT.ai, which allow users to "teach" their personal ChatGPT models easily, but the implementation details are unclear.
Questions:
- What's the best practice for implementing this feature at scale, especially for hundreds of accounts each requiring a customized ChatGPT model?
- Can GPT-3.5 be fine-tuned for individual accounts in a scalable way? If so, how can these models be efficiently managed and utilized?
- Are there any examples or case studies of similar implementations that can guide the development process?