How to Build an Employee Knowledge Base with AI

Employees ask the same questions over and over. Where is the expense policy? How do I request time off? What is the VPN setup? I built an ai employee knowledge base that answers these questions instantly by searching your internal documents and returning the exact answer.

This internal wiki system uses embeddings to find relevant content and AI to generate clear answers.

What You Need Before Starting

• Python 3.8+
• An AI API key (Claude for answers, OpenAI for embeddings)
• Your internal documents (SOPs, policies, guides) in text format
• A vector database (ChromaDB works locally)

Step 1: Collect and Chunk Your Documents

from pathlib import Path

def load_documents(docs_dir):
    documents = []
    for filepath in Path(docs_dir).rglob("*.md"):
        content = filepath.read_text(encoding="utf-8")
        chunks = chunk_text(content, chunk_size=500)
        for i, chunk in enumerate(chunks):
            documents.append({
                "id": f"{filepath.stem}_{i}",
                "text": chunk,
                "source": filepath.name
            })
    return documents

def chunk_text(text, chunk_size=500):
    words = text.split()
    chunks = []
    for i in range(0, len(words), chunk_size):
        chunks.append(" ".join(words[i:i + chunk_size]))
    return chunks

Step 2: Build the Vector Index

import chromadb

def build_index(documents):
    client = chromadb.Client()
    collection = client.create_collection("knowledge_base")

    for doc in documents:
        collection.add(
            documents=[doc["text"]],
            metadatas=[{"source": doc["source"]}],
            ids=[doc["id"]]
        )

    return collection

Step 3: Search for Relevant Content

def search_knowledge(collection, query, top_k=3):
    results = collection.query(
        query_texts=[query],
        n_results=top_k
    )

    context_parts = []
    sources = set()
    for doc, metadata in zip(results["documents"][0], results["metadatas"][0]):
        context_parts.append(doc)
        sources.add(metadata["source"])

    return {
        "context": "\n\n".join(context_parts),
        "sources": list(sources)
    }

Step 4: Generate Answers with AI

import anthropic

ai_client = anthropic.Anthropic()

def answer_question(question, search_results):
    prompt = f"""Answer this employee question using ONLY the context provided.
If the context does not contain the answer, say "I could not find that in our documentation."
Be specific. Include any relevant steps, links, or contacts mentioned.

Context:
{search_results['context']}

Question: {question}

Sources used: {', '.join(search_results['sources'])}"""

    message = ai_client.messages.create(
        model="claude-sonnet-4-20250514",
        max_tokens=1024,
        messages=[{"role": "user", "content": prompt}]
    )
    return {
        "answer": message.content[0].text,
        "sources": search_results["sources"]
    }

Step 5: Wire It Up as a Slack Bot

from slack_bolt import App

app = App(token="xoxb-your-bot-token")

@app.message("kb:")
def handle_kb_query(message, say):
    query = message["text"].replace("kb:", "").strip()
    search_results = search_knowledge(collection, query)
    response = answer_question(query, search_results)

    say(f"*Answer:* {response['answer']}\n\n_Sources: {', '.join(response['sources'])}_")

What to Build Next

Add a feedback button so employees can flag wrong or incomplete answers. Use those flags to identify documentation gaps. When enough people ask about something that is not in the docs, that is your signal to write new documentation.

Related Reading

• Input, Process, Output: The Universal AI Framework - the pattern behind every AI system
• Build vs Buy: The AI Framework - when to build custom vs use off-the-shelf
• The Compounding Advantage Nobody Talks About - how knowledge bases compound over time