Overview
Modern enterprises accumulate massive, fragmented volumes of knowledge across disconnected systems like Confluence pages, Word documents, Jira tickets, Salesforce records, and shared drives. This fragmentation makes it nearly impossible for employees, especially sales and support teams, to quickly find the right answers.
Acheron built an AI-powered Knowledge Discovery Chatbot to unify access to these knowledge silos through a natural, conversational interface. The solution enables employees to get contextual, reference-backed answers drawn from multiple enterprise systems, all within their familiar workspace.
Challenges
Enterprise productivity was significantly hampered by the dispersal of crucial knowledge across multiple tools and formats:
Dispersed Knowledge: Essential information was siloed: Salesforce held deal history; Confluence contained product knowledge and policies; Jira documented implementation details; and shared drives stored contracts and datasheets.
Context Switching: Finding accurate information required employees to manually switch between systems, use varying query structures, and cross-reference documents.
Slower Productivity: This manual search process slowed down query resolution for sales and support teams.
Dependency on Tribal Knowledge: New hires, in particular, struggled to locate relevant information quickly, increasing dependency on senior employees and their tribal knowledge
Objectives
The primary goal of the project was to transform scattered enterprise knowledge into an intelligent, interactive network, specifically to:
Unify Knowledge Access: Create a single, conversational interface to query data from Confluence, Salesforce, Jira, and OneDrive.
Enable Context-Aware Retrieval: Allow employees to ask natural language questions and receive accurate, contextually relevant answers, regardless of the source system.
Improve Efficiency: Significantly reduce the time sales and support teams spend searching for information.
Ensure Transparency: Provide references and source links with every AI-generated response to ensure traceability and trust.
Reduce Dependency: Decrease the reliance on manual support and senior staff for knowledge discovery
How It Works
The system operates based on a Retrieval-Augmented Generation (RAG) architecture:
Ingestion & Embedding: Documents from all connected systems (Confluence, Jira, etc.) are chunked and converted into vector embeddings. These vectors are stored in the ChromaDB vector store.
User Query: An employee enters a question (e.g., “Show me how a similar deal was handled previously”) into the Salesforce-embedded chatbot.
Semantic Search (Retrieval): The query is also converted into a vector and used to search the ChromaDB. The system retrieves the most semantically similar text chunks from the various enterprise documents, even if the exact keywords weren’t used.
Context Generation: The retrieved text chunks, which include source information, are passed to the OpenAI GPT-3.5 model as the context.
Response Generation: GPT-3.5 generates a clear, summarized, and human-like answer based only on the provided context.
Transparency: The final response is delivered with references and source links back to the original documents, ensuring transparency and traceability.
Solution
Acheron developed an AI-driven enterprise chatbot capable of querying multiple enterprise systems and intelligently summarizing the results into clear, actionable responses.
Key Features:
Unified Knowledge Access: Connects directly with Confluence, Salesforce, Jira, and OneDrive, centralizing search into one interface.
Conversational Intelligence: Leverages OpenAI GPT-3.5 to power natural language comprehension and generate human-like, detailed, and appropriate responses.
Context-Aware Retrieval: Uses semantic search based on document embeddings to understand the meaning of a query, not just keywords.
Native Salesforce Integration: The chatbot UI is directly embedded within Salesforce, allowing sales representatives to get contextual answers on the fly. For example:
“What’s the discount limit for this product in Germany?”
“What compliance checks are required before selling in France?”
Learning Feedback Loop: Unanswered queries are routed to Subject Matter Experts (SMEs). The resolved answer is then automatically added to the knowledge base for future recall, ensuring continuous improvement.
Technology Stack
Results
The AI-powered Knowledge Chatbot delivered significant, measurable improvements in employee productivity and organizational consistency:
60% faster query resolution for sales and support teams, directly reducing operational overhead.
Consistent, source-verified responses across multiple systems, standardizing information quality.
Reduced dependency on manual support for knowledge discovery, freeing up senior employees’ time.
Accelerated onboarding for new team members, as they could quickly access relevant information without relying on tribal knowledge.
