Introducing Command R+: Our new, most powerful model in the Command R family.
CHALLENGE
A global network security company wanted to build an AI assistant to help its IT professionals respond to queries on network security setup and security best practices. Its security policy documentation repository includes tens of thousands of articles consisting of unstructured data, and its legacy search tools often produced inaccurate results. The company initially piloted a proof-of-concept (POC) using a competitor’s models, but their high latency meant that generated responses were too slow to meet user testing requirements.
CHALLENGE
A global network security company wanted to build an AI assistant to help its IT professionals respond to queries on network security setup and security best practices. Its security policy documentation repository includes tens of thousands of articles consisting of unstructured data, and its legacy search tools often produced inaccurate results. The company initially piloted a proof-of-concept (POC) using a competitor’s models, but their high latency meant that generated responses were too slow to meet user testing requirements.
SOLUTION
The customer built the AI assistant with retrieval-augmented generation (RAG). Using Cohere Embed for semantic search to generate embeddings of the security policy documentation and Cohere Rerank to further improve retrieval accuracy by selecting the most relevant documents for answer generation. The team then fine-tuned Cohere Command to optimize how it answered user questions. The entire solution delivered the accuracy required at 50% lower latency than previous trials.
SOLUTION
The customer built the AI assistant with retrieval-augmented generation (RAG). Using Cohere Embed for semantic search to generate embeddings of the security policy documentation and Cohere Rerank to further improve retrieval accuracy by selecting the most relevant documents for answer generation. The team then fine-tuned Cohere Command to optimize how it answered user questions. The entire solution delivered the accuracy required at 50% lower latency than previous trials.
How it works
STEP 1.
Unstructured policy documents are embedded by Embed and stored in a vector database.
STEP 2.
Command interprets a user’s requests and creates queries to retrieve relevant answers.
STEP 3.
Rerank re-orders the retrieved answers based on relevance to original queries, improving the accuracy of the search results.
STEP 4.
Command synthesizes a conversational response back to the user, incorporating answers along with citations.
Impact
Higher user satisfaction
Less effort required by users to find answers
Accurate answers to domain-specific questions
The Cohere Difference
Leading model accuracy
Cohere’s retrieval prioritizes accurate responses and citations
Accelerated enterprise deployment
Cohere’s models come with connectors to common data sources
Customization
Cohere’s models can be fine-tuned to further improve domain performance
Scalability
Cohere’s powerful inference frameworks optimize throughput and reduce compute requirements
Flexible deployment
Cohere models can be accessed through a SaaS API, on cloud infrastructure (Amazon SageMaker, Amazon Bedrock, OCI Data Science, Google Vertex AI, Azure AI), and private deployments (virtual private cloud and on-premises)
Multilingual support
Over 100 languages are supported, so the same topics, products, and issues are identified the same way