Business Context
TechSupport Inc. is a SaaS company that provides cloud solutions to businesses. They receive approximately 1,000 customer inquiries daily through various channels, including email and live chat. The customer support team is overwhelmed, leading to long response times and decreased customer satisfaction. They want to implement a chatbot powered by a Large Language Model (LLM) to handle frequently asked questions (FAQs) and reduce the workload on human agents.
Data Characteristics
- Volume: 50,000 historical customer inquiries labeled with intents and responses.
- Text Length: Average of 20-100 words per inquiry.
- Language: English.
- Label Distribution: 30% billing inquiries, 25% technical support, 20% account management, 25% general questions.
Success Criteria
- Achieve 90% accuracy in intent recognition.
- Reduce average response time from 10 minutes to under 5 minutes.
- Ensure at least 70% of inquiries can be resolved without human intervention.
Constraints
- The model must comply with data privacy regulations (e.g., GDPR).
- The solution should integrate seamlessly with existing support ticketing systems.
- Must be deployable on cloud infrastructure with cost-effective scaling.
Requirements
- Fine-tune an LLM (e.g., GPT-3) on the customer inquiry dataset.
- Implement a preprocessing pipeline for user inquiries.
- Develop a user-friendly interface for customers to interact with the chatbot.
- Set up monitoring to track performance and user satisfaction.
