Enterprise Knowledge

Most AI platforms introduce a concept called a Knowledge Base.

The workflow usually looks like this:

Upload Documents
        ↓
Knowledge Base
        ↓
Chatbot

While simple, this approach often becomes difficult to scale in enterprise environments.

Organizations rarely have a single knowledge base.

Instead they have:

• HR policies
• IT documentation
• Engineering runbooks
• Legal contracts
• Financial procedures
• Customer support documentation

Each domain has different ownership, governance, security requirements, and retrieval patterns.

---

The Karyam Approach

Karyam treats knowledge as infrastructure rather than a standalone object.

Instead of creating “Knowledge Bases”, organizations build retrieval pipelines using:

Folders
    +
Embedding Listeners
    +
Embedding Models
    +
Vector Databases
        ↓
Enterprise Knowledge

This architecture provides significantly greater flexibility and scalability.

---

Traditional Knowledge Bases

Traditional systems often look like:

HR Knowledge Base
    ├── hr-policy.pdf
    ├── leave-policy.pdf
    └── handbook.pdf

IT Knowledge Base
    ├── vpn-policy.pdf
    └── access-sop.pdf

This works well initially but becomes difficult when:

• Multiple teams share content
• Security requirements differ
• Documents need different embedding strategies
• Multiple vector databases are required
• Different retrieval pipelines are needed

---

Knowledge Organization In Karyam

In Karyam, the same structure looks like:

knowledge/
│
├── hr/
│   ├── employee-handbook.pdf
│   ├── leave-policy.pdf
│   └── benefits-guide.pdf
│
├── finance/
│   ├── procurement-policy.pdf
│   └── reimbursement-policy.pdf
│
├── engineering/
│   ├── architecture.md
│   └── runbooks/
│
└── it/
    ├── vpn-policy.pdf
    └── access-sop.pdf

Each folder can have its own:

• Embedding listener
• Embedding model
• Vector database
• Metadata rules
• Access controls

---

Example Architecture

HR Folder
    ↓
HR Listener
    ↓
HR Vector Database

IT Folder
    ↓
IT Listener
    ↓
IT Vector Database

Engineering Folder
    ↓
Engineering Listener
    ↓
Engineering Vector Database

This enables domain-specific retrieval strategies.

---

Advantages Of This Model

Independent Scaling

Engineering documentation may require millions of vectors while HR documentation may only require thousands.

Each domain can scale independently.

---

Flexible Embedding Models

Different knowledge domains benefit from different embedding models.

Example:

HR Documents
    ↓
OpenAI Embeddings

Source Code
    ↓
Voyage Code Embeddings

---

Independent Governance

Different teams can own different retrieval pipelines.

Example:

Domain	Owner
HR	HR Team
Finance	Finance Team
Engineering	Platform Team
Legal	Compliance Team

---

Multiple Vector Databases

Organizations can use:

HR
↓
PGVector

Engineering
↓
Qdrant

depending on workload requirements.

---

Example: IT Support Knowledge

Suppose we want to build an IT Support Assistant.

We create:

knowledge/it/
├── vpn-policy.pdf
├── access-sop.pdf
├── onboarding-guide.pdf
└── security-guidelines.pdf

Configure:

Folder:
knowledge/it

Listener:
IT Embedding Listener

Embedding Model:
text-embedding-3-large

Vector Database:
IT Support DB

Every uploaded file automatically becomes searchable.

---

How Agents Use Knowledge

Agents never access files directly.

Instead:

User Question
        ↓
Query Embedding
        ↓
Vector Search
        ↓
Relevant Chunks Retrieved
        ↓
Agent Response

This allows the same knowledge infrastructure to serve:

• Agents
• Flows
• Automations
• Internal search systems

---

Shared Knowledge Infrastructure

One retrieval pipeline can power multiple systems.

Example:

IT Knowledge
        ↓
├── IT Support Agent
├── Employee Assistant
├── Access Approval Flow
└── Internal Search Portal

This avoids duplication and improves consistency.

---

Knowledge Is Infrastructure

Karyam views enterprise knowledge as infrastructure rather than application configuration.

This approach enables:

• Shared context
• Reusable retrieval pipelines
• Better governance
• Independent scaling
• Production observability

---

The Karyam Model

Files
    ↓
Listeners
    ↓
Embeddings
    ↓
Vector Databases
    ↓
Retrieval
    ↓
Agents & Flows

Knowledge is not a feature.

It is a platform capability.

---

Next Steps

You now understand the complete knowledge architecture of Karyam:

• Vector Databases
• Embeddings
• File Uploads
• Chunking
• Retrieval
• RAG Runs
• Knowledge Organization

Together these components form the foundation of enterprise AI systems.

---