- Newest
- Most votes
- Most comments
In the scenario you’ve outlined where Amazon Q Connect offers assistance for a complex query but struggles with simpler ones, the issue likely stems from how the Large Language Model (LLM) interprets customer utterances and matches them to predefined intents or knowledge base content.
Here's what can be done to improve LLM response for simpler utterances like “I am a new patient” or “I need an appointment”:
-
Refining LLM Prompts for Simpler Queries Amazon Q in Connect now allows customization of the LLM prompts through AI Prompts. To better handle simple utterances: • Customize the LLM prompts to focus on identifying common, simpler intents. Use tailored instructions to reformulate customer queries such as "I need an appointment" into clear, actionable search queries. • Use text completions format to direct the LLM to pull relevant knowledge base excerpts for simpler requests, ensuring that even minimal context triggers helpful responses.
-
Intent Mapping Simpler queries might not trigger a specific intent due to vague or broad phrasing. To fix this: • Expand the lexicon of supported intents in Amazon Lex or other connected intent recognition systems to handle variations of basic requests. This helps the system better capture “I need an appointment” or “I am a new patient” as meaningful actions. • Incorporate conversation transcripts and flow menus from Amazon Connect into session data, which can guide the LLM to interpret the utterances correctly.
-
Utilize Amazon Connect Customer Profiles for Contextual Understanding When someone says, “I am a new patient,” Amazon Q can integrate customer profile data from Amazon Connect to provide a personalized response: • Link customer data sources (like CRM or patient management systems) so that even simple queries result in personalized responses. For example, “I need an appointment” could trigger an automatic response based on the customer’s profile (new vs. returning patient, appointment history, etc.).
-
Enhancing Query Reformulation • Amazon Q in Connect can now use Query Reformulation AI prompts to rephrase vague queries into more specific ones that can trigger useful responses. For instance, a query like "I need an appointment" can be reformulated into "What are the available appointment slots for new patients?" to improve knowledge base search results.
-
Testing and Iterating Custom Solutions • Implement custom AI Agents to manage how different prompts and behaviours apply to simple vs. complex queries. This ensures that regardless of the complexity, the LLM always provides a relevant answer. • Test and iterate the response patterns for simpler queries by simulating customer interactions and adjusting the AI Prompts and Intent mapping where necessary.
By focusing on refining LLM prompts, intent recognition, and leveraging customer data, you can ensure that Amazon Q in Connect offers the right help for both simple and complex customer utterances, leading to a more responsive and intuitive customer service experience.
Relevant content
- asked 3 months ago
- asked 2 years ago
- AWS OFFICIALUpdated a year ago
- AWS OFFICIALUpdated 2 years ago
- AWS OFFICIALUpdated 2 years ago
- AWS OFFICIALUpdated 2 years ago