Great — now let’s train your AI agent on company-specific leads, services, and response styles so it answers like your business would, not just like a generic assistant.

🧠 Lesson: Training Your Agent on Company-Specific Leads, Services & Responses

🎯 Lesson Objective

By the end of this lesson, you’ll be able to:

  • Provide your AI agent with deep knowledge of your products, services, offers, and lead types

  • Teach it how your company communicates (tone, structure, policies)

  • Make it respond with branded, accurate, and confident messaging

  • Store and retrieve company-specific information using RAG (Retrieval-Augmented Generation)

🧠 Why Train on Your Own Data?

Generic LLMs are powerful — but they:

  • ❌ Don’t know your pricing, services, or guarantees

  • ❌ Use generic tone, which may clash with your brand

  • ❌ Can’t answer specific lead objections unless trained

Training your agent on company data helps it:

✅ Handle real client questions
✅ Stay consistent with your brand voice
✅ Answer quickly without hallucinating

🧰 Tools You’ll Use

Tool Purpose
LangChain Framework for RAG and chaining
Ollama / OpenAI LLM backend
Chroma / FAISS Vector DB to store internal data
txt / PDF / CSV loaders Load your company content
PromptTemplate Brand-style responses

🔧 Step-by-Step: Training Your Agent

✅ 1. Gather Your Company Data

Collect all sources:

  • Product/service descriptions

  • Price lists

  • FAQ & customer service scripts

  • Past email/chat examples

  • Sales objections & how you handle them

Store in text files (.txt, .pdf, .docx, .md)

✅ 2. Load & Chunk the Data

from langchain.document_loaders import PyPDFLoader, TextLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter

# Load documents
loader = TextLoader("company_services.txt")  # Or PyPDFLoader("pricing.pdf")
documents = loader.load()

# Split into chunks
splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=50)
docs = splitter.split_documents(documents)

✅ 3. Create a Vector Store (Knowledge Base)

from langchain.vectorstores import Chroma
from langchain.embeddings import OllamaEmbeddings  # Or OpenAIEmbeddings

embeddings = OllamaEmbeddings(model="nomic-embed-text")
vectorstore = Chroma.from_documents(docs, embedding=embeddings)

✅ 4. Add Retrieval QA to Your Agent

from langchain.chains import RetrievalQA
from langchain.llms import Ollama

llm = Ollama(model="mistral")

qa_chain = RetrievalQA.from_chain_type(
    llm=llm,
    retriever=vectorstore.as_retriever(),
    return_source_documents=True
)

query = "What’s the difference between our Basic and Premium plans?"
result = qa_chain.run(query)
print("Agent:", result)

✅ 5. Teach Brand Voice with PromptTemplate

from langchain.prompts import PromptTemplate

prompt = PromptTemplate(
    input_variables=["question", "context"],
    template="""
You are a friendly, helpful customer assistant for GreenTech Solutions.

Answer the following question using only the information provided. Keep the tone {tone} and concise.

Context:
{context}

Question:
{question}

Answer:
"""
)

Plug this into your retrieval agent’s chain to shape tone and format.

🧪 Example Queries It Can Now Answer:

Query Agent Answer (Trained)
“How long is your warranty?” “We offer a 24-month warranty on all solar tracking units.”
“Do you provide financing?” “Yes, we partner with GreenBank to offer low-interest financing options.”
“Why choose you over [competitor]?” “Unlike many competitors, we include installation, calibration, and free support for 1 year.”

🧠 Enhance It Further

Feature Description
🎯 Lead-specific replies Store industry or client type in memory and tailor response
📂 Multi-file indexing Combine PDFs, web copy, and past chats
🔁 Dynamic updates Re-index every time your services change
🌐 Use by sales or support teams Integrate with Streamlit or chatbots

✅ Summary

Step What You Did
🗃️ Gathered company-specific data Services, policies, pricing, tone
📚 Created a knowledge base Via vector store & embeddings
💬 Built an AI assistant That responds accurately and consistently
🧠 Trained it on real company language Using prompt templates + documents

🔜 Want to Go Deeper?

Would you like to:

  1. 🧾 Add live updates from your website or Notion page?

  2. 👥 Build a multi-user dashboard for sales/support teams?

  3. 🤖 Create a chatbot version of this trained agent for customers?

 

49
PreviousNext