ChatFactory: A Smarter Buyer Agent

Written by: Srinivasan Margabandhu, Executive Director, AI Research

This year marks a pivotal period for Agent-based systems transforming from a promising technology into a game-changing business reality. Thanks to revolutionary advances in Large Language Models, today’s AI agents deliver unprecedented intelligence, sophisticated reasoning, & effective intent understanding capabilities.

However, here’s the key insight: raw intelligence alone isn’t enough. Just like humans, AI agents require the right context to leverage their capabilities and deliver business-driving results. That’s where Context Engineering plays an important role in Generative AI solutions like ChatFactory.

ChatFactory Buyer Agent 

PathFactory has unleashed major enhancements to ChatFactory. In this article, we will reveal the major breakthrough techniques and methodologies that are already delivering impact for forward-thinking businesses.

In ChatFactory, the ability to provide accurate responses to buyer queries is paramount. This leads to several key benefits:

• Enhanced B2B Buyer Experience: Precise, context-driven answers reduce friction and improve the overall buyer experience. It enables a guided experience to buyers who know best about their problem but not the product.
• Operational Efficiency: High-relevance retrieval minimizes irrelevant information, boosting productivity and accelerating decision-making.
• Brand Credibility: Delivering factual and consistent responses builds trust, particularly when buyers depend on the platform for accurate information.

Not all buyer queries are equal 

Inspired by today’s evolving landscape, buyers are asking increasingly sophisticated questions, ranging from highly-focused inquiries to complex generic ones. This necessitates a more nuanced approach to providing answers, as not all buyer questions are equal.

The core to delivering effective responses lies in the use of multiple Retrieval Augmented Generations (RAGs), which should adapt to the diverse nature of these queries. Understanding both the individual query intent and the overall buyer intent is crucial.

Often, the answers to these complex questions are not found in a single location; they may be dispersed across different sections of one content piece or even across multiple pieces of different content. Furthermore, AI systems must be able to connect different entities to form a comprehensive understanding of the full picture.

Ultimately, this is a “context engineering” problem. It involves engineering the most relevant context to the Large Language Model (LLM), ensuring that responses are not only accurate but also clear and concise.

The above diagram illustrates the spectrum of approaches, all working towards context engineering, showing how methods are adaptive from basic, direct retrieval to more sophisticated methods like GraphRAG. As you move toward the top-right, context engineering assumes more significance – enabling RAG systems to deliver increasingly precise, relevant, and nuanced answers by tapping into structured knowledge, multi-step reasoning, intent detection, and dynamic context assembly.

Graph RAG

ChatFactory’s RAG systems, like fine wine, get better with time. Graph RAG, a recent addition, integrates content knowledge graphs into the retrieval and generation loop. It works on the principle of discovering entities (as nodes) from contents, extracting relationships among them, & vectorizing these relationships as edges within a graph. So all related contents are connected. The advantage of this approach is to bring together nuanced relationships that exist between entities.

For example, consider a buyer query like this: How does ABM work for you and bring out customer references for Pharma. The answer could come from multiple documents. In this case a content on ABM plus another set of contents signifying customer stories on ABM, blended with a case study from Healthcare customers. These relationships are best connected online from a Knowledge Graph which is built, pre-curated for all contents within a pre-defined content pool. Simple semantic search from traditional Graph RAG approaches will not be able to succinctly identify the relevant contents. It is here that the Graph RAG agent will be useful.

A hybrid search option

Our RAG incorporates proven search/retrieval methodologies, like BM25, to retrieve contexts based on buyer query. The platform employs entity & keyword extraction to parse buyer queries, subsequently applying multi-tiered filtering mechanisms that operate at both page-level and chunk-level granularity to ensure intent-content alignment. This approach is operationalized through our Buyer Intent RAG implementation, which leverages hybrid search capabilities across our enterprise vector database infrastructure to deliver contextually relevant results. The diagram illustrates this concept with an example.

Ensemble RAG: How it all fits together

Now that we have the individual AI RAG agents working together, an Ensemble Agent orchestrates them, assembles the contexts, then reranks to further fine tune the context before interacting with the LLM. Here, the reranker is a sort of a noise-cancellation system applied to the RAG outputs. They are represented in the following diagram.

The optimized context finally is passed on to the LLM, along with the buyer query to craft a personalized response.

Conclusion

By strategically engineering context and leveraging advanced RAG techniques, ChatFactory’s Buyer Agent now is able to deliver a more accurate and relevant response, continuously evolving to meet the complex demands of modern buyer interactions.