esynergy Prepares Proposals 75% Faster with AI Document Retrieval from AWS and DataStax

esynergy leverages a community of subject matter experts to design and implement technology solutions that support business objectives. Known for its problem-solving approach to industry challenges, esynergy wanted to build a solution to improve the document retrieval process for customers—and for its own sales team.

“Over the past 10 years, our teams have generated numerous documents stored in various formats within SharePoint, such as PowerPoint proposals, project narratives, visual artifacts and PDF case studies,” said Prasad Prabhakaran, generative AI practice lead at esynergy. “When customers ask for information on practices like migrating workloads to the cloud, our sales representatives often must consult multiple sources and navigate through a vast array of documents on SharePoint, which can take days. We knew many of our customers were dealing with similar issues.”

Existing AI solutions failed to meet esynergy’s requirements for context awareness, security, and scalability. The company needed a cost-effective, scalable database to handle large volumes of data and provide rapid, accurate search results using AI. The solution also had to integrate seamlessly with existing tools, such as SharePoint for document storage and Amazon Web Services (AWS).

To address the challenges faced by its sales teams, esynergy developed Sales Copilot, an AI-powered conversational assistant that enables rapid access to relevant customer data and generates informed responses. The solution leverages DataStax Astra DB, a scalable and cost-effective vector database, to handle large volumes of data and provide fast, accurate search results using AI.

"We chose Astra DB for its serverless architecture, which significantly reduced our operational costs," said Prabhakaran. "Additionally, its caching capabilities and ability to manage multiple turnarounds to the large language model (LLM), Claude 2, allowed us to deliver fast and accurate responses to our users, even under high concurrency. These features made Astra DB the perfect fit for our Sales Copilot application."

The Sales Copilot application seamlessly integrates with the organization's SharePoint collaboration platform, ingesting customer profiles, communication records, and documents like case studies. LangChain, a framework for building applications with LLMs, plays a crucial role in the data preprocessing stage, enabling efficient splitting of files into manageable chunks for subsequent processing. These chunks are then transformed into dense vector representations using embeddings generated by Astra DB, which are optimized for fast and accurate retrieval. The "retrieve-augment" architecture relies on esynergy's customized component to return only the most relevant data points from Astra DB.

The Claude 2 LLM, integrated through Amazon Bedrock, crafts relevant responses tailored to the user's query, informed by the retrieved customer data and documents. The modular components of LangChain make the implementation of this architecture straightforward, allowing for rapid prototyping and experimentation with different strategies. “Along with Astra DB, Amazon Bedrock provided us with the flexibility to experiment with different AI models and find the one that best suited our needs,” said Prabhakaran. “It allowed us to easily integrate Claude 2 into our solution, providing powerful natural language processing capabilities without the hassle of managing the underlying infrastructure.”

The Sales Copilot application integrates seamlessly with esynergy's existing AWS infrastructure, including AWS PrivateLink for enhanced security and compliance. By leveraging the easy-to-use features of Astra DB, Amazon Bedrock, and LangChain, esynergy was able to develop and deploy the Sales Copilot application within a short timeframe, enabling rapid prototyping and iteration based on user feedback and changing requirements.