Orchestrating Multi-Agent RAG Pipelines & LLMOps for Enterprise Operations

// Layer 3: Retrieval Orchestration (LangChain)
import { ChatOpenAI } from '@langchain/openai'
import { PGVectorStore } from '@langchain/community/vectorstores/pgvector'
import { OpenAIEmbeddings } from '@langchain/openai'
import { createRetrievalChain } from 'langchain/chains/retrieval'
import { createStuffDocumentsChain } from 'langchain/chains/combine_documents'
import { ChatPromptTemplate } from '@langchain/core/prompts'

const embeddings = new OpenAIEmbeddings({ model: 'text-embedding-3-large' })

const vectorStore = await PGVectorStore.initialize(embeddings, {
  postgresConnectionOptions: { connectionString: process.env.DATABASE_URL },
  tableName: 'knowledge_vectors',
  columns: {
    idColumnName: 'id',
    vectorColumnName: 'embedding',
    contentColumnName: 'content',
    metadataColumnName: 'metadata',
  },
})

// CRITICAL: System prompt guardrails — prevents hallucination
const systemPrompt = `You are a specialized assistant for {company_name}.
Answer ONLY using the provided context documents.
If the answer is not found in the context, respond with:
"I don't have verified data for this query. Please contact the operations team."
NEVER fabricate figures, names, or policies.

Context:
{context}`

const prompt = ChatPromptTemplate.fromMessages([
  ['system', systemPrompt],
  ['human', '{input}'],
])

const llm = new ChatOpenAI({ 
  model: 'gpt-4o',
  temperature: 0,      // Zero temperature = deterministic, factual responses
  maxTokens: 1024,
})

const questionAnswerChain = await createStuffDocumentsChain({ llm, prompt })
const ragChain = await createRetrievalChain({
  retriever: vectorStore.asRetriever({ k: 6 }),  // Retrieve top 6 chunks
  combineDocsChain: questionAnswerChain,
})

// Query the RAG chain
const result = await ragChain.invoke({
  input: 'What is the current stock level for SKU-MT-4422?',
  company_name: 'Fabrication Plant Darbhanga',
})

console.log(result.answer)  // Grounded, source-cited response

// Router Agent — classifies query intent and routes to specialist
import { ChatOpenAI } from '@langchain/openai'

type AgentRoute = 'inventory' | 'finance' | 'customer_service' | 'hr' | 'unknown'

async function routerAgent(query: string): Promise<AgentRoute> {
  const router = new ChatOpenAI({ model: 'gpt-4o-mini', temperature: 0 })
  
  const response = await router.invoke([
    {
      role: 'system',
      content: `Classify the following query into one of these categories:
inventory, finance, customer_service, hr, unknown.
Respond with ONLY the category name — nothing else.`
    },
    { role: 'user', content: query }
  ])
  
  return response.content as AgentRoute
}

// Multi-agent dispatcher
async function dispatchQuery(userQuery: string, userId: string) {
  const route = await routerAgent(userQuery)
  
  // Route to specialist agent
  switch (route) {
    case 'inventory':
      return inventoryAgent.invoke({ input: userQuery, userId })
    case 'finance':
      return financeAgent.invoke({ input: userQuery, userId })
    case 'customer_service':
      return customerServiceAgent.invoke({ input: userQuery, userId })
    default:
      return { answer: 'This query requires human assistance. Routing to operations team.' }
  }
}