🌿💻 The Packets Garden

Amazon Kendra

🌱 🪴 ⏱️ 4 min read

Main Concept

Amazon Kendra is a fully managed document search service powered by ML. It indexes documents from multiple sources and allows users to ask natural language questions — returning direct, precise answers extracted from within the documents rather than a list of links to search through.

Key Idea

  • Input → a natural language question from a user.

  • Output → a direct, precise answer extracted from indexed internal documents.

  • Exam trigger → any scenario mentioning “document search” or “intelligent search” → Amazon Kendra.

Supported Document Types

Kendra indexes and searches across a wide range of document formats:

Text, PDF, HTML, PowerPoint, Microsoft Word, FAQs, and more.

Traditional search  → user searches a keyword → returns a list of documents
                      user must open and read each document to find the answer

Amazon Kendra       → user asks "Where is the IT support desk?"
                      Kendra returns "1st floor"
                      direct answer extracted from the indexed documents

Example from Maarek's lesson

User asks: “Where is the IT support desk?” Kendra searches across all indexed internal documents. Kendra responds: “1st floor.”

Kendra found the answer buried in an internal document and surfaced it directly — without the user needing to know which document contained it or even that the document existed.

Key Features

Users ask questions in plain natural language — no need to know keywords, document names, or where information is stored. Kendra understands the question and finds the answer.

Incremental Learning

Kendra learns from user interactions and feedback over time — promoting search results that users found helpful and demoting ones they ignored. Search quality improves continuously without manual retraining.

Key Idea

  • Incremental learning → Kendra gets smarter the more it is used, based on real user behavior and feedback.

Fine-Tuned Search Results

Search results can be customized based on:

Data importance   → prioritize certain document types or sources
Freshness         → surface more recent documents first
Custom filters    → apply business-specific rules to result ranking

Common Exam Scenarios

Key Idea: When the answer is Amazon Kendra

  • “Employees need to search across thousands of internal documents to find policy information” → Amazon Kendra.

  • “Build an intelligent FAQ system that answers questions from internal documentation” → Amazon Kendra.

  • “A company wants natural language search across their knowledge base instead of keyword search” → Amazon Kendra.

  • “Surface direct answers from documents instead of returning a list of links” → Amazon Kendra.

  • Any scenario involving intelligent document search or enterprise knowledge search → Amazon Kendra.

Critical Distinctions

Amazon Kendra      → SEARCHES documents to find direct answers
                     understands natural language questions
                     returns precise answers from indexed repositories

Amazon Comprehend  → UNDERSTANDS text meaning
                     sentiment, entities, topics
                     does not search across documents

Amazon Lex         → BUILDS conversational chatbots
                     manages dialogue and intent
                     can integrate WITH Kendra to answer questions

Kendra + Lex together

A company builds an internal HR chatbot using Amazon Lex. When an employee asks “How many vacation days do I have left?” Lex handles the conversation and invokes Kendra to search the HR policy documents for the answer — then returns it to the user through the chatbot interface.

Exam Scope

You will not be asked how to implement Kendra. You need to:

  • Know what Kendra does (intelligent document search — natural language questions, direct answers).
  • Know the key differentiator from traditional search (direct answers vs list of documents).
  • Know the supported document types (text, PDF, HTML, PowerPoint, Word, FAQs).
  • Know incremental learning as a feature (improves from user feedback automatically).
  • Distinguish Kendra (search documents) from Comprehend (understand text) and Lex (build chatbots).
  • Maarek’s own exam tip: “whenever you see a document search service, think Amazon Kendra.”

Exam Domain

  • Domain 1, Task Statement 1.2: “Explain the capabilities of AWS managed AI/ML services (for example, Amazon Kendra).”
  • Domain 1, Task Statement 1.2: “Identify examples of real-world AI applications (for example, search).”

Graph View

Backlinks