Amazon Kendra is an AI-powered document search service from AWS.
👉 In simple terms:
It lets you index documents from multiple sources into a central repository and enables natural language search over them.
Unlike basic keyword search, Kendra uses ML/NLP to understand intent and return context-aware answers.
📚 1. Kendra as a Document Search Service
Kendra acts like:
“Google for your enterprise documents”
Key capabilities:
Centralized document indexing
Natural language querying
Extracts answers (not just links)
Role-based access filtering
🧠 2. Does it create a central index?
👉 Yes — this is core to Kendra
You create an Index
All documents are ingested into this index
Search queries run against this index
Architecture:
Data Sources → Kendra Index → Search API → Application / UI
📄 3. Supported Document Types
Kendra supports a wide range of formats:
📁 Common formats:
PDF
Word (DOC, DOCX)
Excel (XLS, XLSX)
PowerPoint (PPT, PPTX)
HTML
XML
JSON
Plain text
🧾 Structured + semi-structured:
FAQs
Knowledge base articles
Wiki pages
Emails (via connectors)
🖼️ Images?
Not directly searchable
But can be indexed if:
Text is extracted using:
Amazon Textract
💬 4. Natural Language Search
👉 One of Kendra’s strongest features
Example queries:
“What is the leave policy for contractors?”
“How to reset VPN password?”
“Show SLA for premium customers”
What happens internally:
Query understanding (NLP)
Semantic matching (not just keywords)
Ranking based on relevance
👉 Output:
Direct answers (highlighted)
Ranked documents
🔗 5. Integrations (Very Powerful)
Kendra integrates with many enterprise systems:
📦 AWS-native sources:
Amazon S3
Amazon RDS
Amazon DynamoDB
🏢 SaaS / enterprise tools:
SharePoint
OneDrive
Google Drive
Confluence
Salesforce
ServiceNow
👉 (via built-in connectors)
🔌 Custom sources:
Use:
Kendra APIs
Custom connectors
🖥️ 6. How to Use from AWS Console
Step-by-step:
1️⃣ Create Index
Go to Kendra → Create index
Configure:
Name
IAM role
Capacity
2️⃣ Add Data Sources
Choose connector:
S3 / SharePoint / etc.
Configure access
Start sync
3️⃣ Indexing
Documents are:
Crawled
Parsed
Indexed
4️⃣ Search
Use:
Console search UI
API (
QueryAPI)
5️⃣ Build Application
Integrate search into:
Web apps
Chatbots
Internal tools
🔐 7. Authentication & Security
Kendra supports multiple auth mechanisms:
🔑 1. IAM (Primary)
Access via:
AWS SDK / CLI
Controlled via IAM roles & policies
🧑💼 2. User Context Filtering
Document-level permissions
Integrated with:
Active Directory
SSO systems
👉 Ensures:
Users only see documents they are allowed to
🌐 3. API Access
Signed requests (SigV4)
Used by applications
🔐 4. Identity Providers
SAML-based SSO
Integration with enterprise identity systems
⚙️ 8. How Kendra Works Internally (Simplified)
Ingestion → Parsing → NLP Enrichment → Indexing → Query Engine
Extracts metadata
Understands document structure
Builds semantic index
🧩 9. Advanced Features
🔹 FAQ support
Direct Q&A matching
🔹 Relevance tuning
Boost certain documents
🔹 Custom metadata
Filter search results
🔹 Incremental sync
Only updates changed documents
🚀 10. Where Kendra Fits (Important Insight)
Given your background (RAG, GenAI, document parsing):
👉 Kendra can replace parts of your pipeline:
Instead of:
Parsing → Chunking → Embedding → Vector DB → Retrieval
You can use:
Kendra Index → Query API → Results
👉 Or combine:
🔥 Kendra + LLM (Best pattern)
Kendra → retrieval
LLM → summarization / reasoning
🏁 Final Summary
Amazon Kendra = intelligent document search engine
Creates a central index
Supports:
Multiple document formats
Natural language queries
Integrates with:
AWS + enterprise tools
Access via:
Console
APIs
IAM / SSO
💬 One-line takeaway:
Kendra is a managed enterprise search + semantic retrieval system, ideal for building internal knowledge search and RAG-style applications.
If you want, I can:
Compare Kendra vs OpenSearch vs Vector DB (very useful for your use case)
Show architecture for Kendra + LLM chatbot
Or suggest when NOT to use Kendra (cost/performance tradeoffs)
No comments:
Post a Comment