2.5 Vector Index Scale, Migration & Retrieval Evaluation

AWS Certified Generative AI Developer - Professional AIP-C01

Track your progress as you read

Finished sections get a check mark.

1 Chapter 1: Exam Overview & Foundation Model Fundamentals

1.1 AIP-C01 Exam Structure, Logistics & Passing Strategy 1.2 AIP-C01 Scoring & Professional Decision Strategy 1.3 Foundation Model Landscape & Model Selection on Amazon Bedrock 1.4 Model Abstraction, Controlled Switching & Resilience 1.5 Customization Spectrum: Prompting vs RAG vs Fine-Tuning vs Continued Pre-training 1.6 Custom Model Deployment & Lifecycle Management

2 Chapter 2: Bedrock Knowledge Bases & RAG Architectures

2.1 Managed RAG Architecture with Bedrock Knowledge Bases 2.2 Knowledge Source Integration, Synchronization & Freshness 2.3 Data Ingestion, Parsing & Chunking Strategies 2.4 Embedding Models & Vector Representations 2.5 Vector Index Scale, Migration & Retrieval Evaluation 2.6 Vector Database Selection: OpenSearch Serverless vs Aurora pgvector vs Managed Stores 2.7 Vector Store Security, Multi-Tenancy & Connectivity

3 Chapter 3: Advanced Vector Search, Indexing & Data Compliance

3.1 Hybrid Search, Metadata Filtering & Semantic Reranking 3.2 Metadata Filtering, Reranking & Query Transformation 3.3 Fine-Tuning & Continued Pre-Training Data Pipeline Engineering 3.4 Data Validation, Multimodal Processing & Model Formatting 3.5 Data Privacy, Data Residency & Compliance Governance 3.6 Privacy Preservation, Source Lineage & Audit Evidence

4 Chapter 4: Prompt Engineering & Bedrock Runtime APIs

4.1 Advanced Prompt Engineering Techniques & Inference Parameters 4.2 Inference Parameters, Prompt Governance & Regression 4.3 Converse API vs InvokeModel API & Message Formatting 4.4 Tool Calling, FM API Resilience & Model Routing 4.5 Streaming Architectures & Asynchronous Event Processing 4.6 Resilient Streaming, Deployment & Asynchronous Processing

5 Chapter 5: Amazon Bedrock Agents & Tool Integration

5.1 Agentic Architecture, ReAct Reasoning Loops & Orchestration 5.2 Agent Frameworks, MCP & Lifecycle Management 5.3 Action Groups, OpenAPI Schemas & AWS Lambda Integration 5.4 Agent Memory, Session State & Multi-Agent Collaboration Patterns 5.5 Persistent State & Multi-Agent Coordination 5.6 Advanced Agent Prompt Templates & Flow Customization 5.7 Agent Safeguards, Tool Reliability & Cost Control

6 Chapter 6: Workflow Orchestration & Application Integration

6.1 Amazon Bedrock Prompt Flows & Visual Orchestration 6.2 Step Functions Orchestration & Human-in-the-Loop Workflows 6.3 Enterprise Integration: API Gateway, AppSync, and EventBridge 6.4 Hybrid Integration, GenAI Gateways & CI/CD Tooling

7 Chapter 7: AI Safety & Guardrails for Amazon Bedrock

7.1 Guardrails Architecture, Denied Topics & Content Filters 7.2 Content Safety, Defense in Depth & Threat Detection 7.3 Sensitive Information Filters & PII Redaction/Masking 7.4 Contextual Grounding Checks & Hallucination Prevention

8 Chapter 8: Security, Access Control & Network Architecture

8.1 IAM Least Privilege & Resource Policies for Bedrock 8.2 PrivateLink VPC Endpoints & Network Isolation 8.3 Encryption with AWS KMS & Cross-Account Access Patterns

9 Chapter 9: Governance, Auditing & Operational Logging

9.1 SageMaker Model Cards & AI Governance Frameworks 9.2 Lineage, Fairness, Transparency & Continuous Governance 9.3 Invocation Logging, CloudTrail Auditing & Observability 9.4 Responsible AI, Bias Mitigation & Red Teaming Methodologies

10 Chapter 10: Performance, Throughput & Cost Optimization

10.1 On-Demand vs Provisioned Throughput & Capacity Planning 10.2 Capacity Measurement, Scaling Patterns & Resilience 10.3 Batch Inference & Bedrock Intelligent Prompt Routing 10.4 Intelligent Prompt Routing & Application Eligibility 10.5 Prompt Caching, Token Optimization & Latency Reduction

11 Chapter 11: Testing, Evaluation & Troubleshooting GenAI Solutions

11.1 Amazon Bedrock Model Evaluation (Automatic & Human Workflows) 11.2 Human, RAG & Continuous Evaluation 11.3 Evaluation Metrics & RAG Quality Benchmarking (Ragas, ROUGE, BLEU) 11.4 RAG and Agent Performance Evaluation 11.5 CloudWatch Metrics, Throttling & Operational Observability 11.6 Quota Diagnosis, Retry Control & Operational Troubleshooting 11.7 Troubleshooting Common GenAI Failure Modes & Agent Traces 11.8 Prompt Maintenance, Content Handling & Deployment Validation

2.5 Vector Index Scale, Migration & Retrieval Evaluation

Key Takeaways Explain This Topic AI

Approximate-neighbor parameters trade recall for latency and memory.

Benchmark restrictive metadata filters because they can materially change search behavior.

Embedding-version changes normally require re-embedding, parallel validation, cutover, and rollback.

Last updated: September 2026

• 2.5 Vector Index Scale, Migration & Retrieval Evaluation
• Scale and index operations

Approximate nearest-neighbor methods trade some recall for speed. HNSW graph parameters affect memory, build time, and search recall. Inverted-file approaches depend on training and probe choices. Managed services hide parts of this machinery, but operators still need to monitor latency, throttling, storage, index health, ingestion lag, and recall on canary queries.

Sharding and multi-index designs should follow access patterns. Separate indices can isolate departments or languages, but excessive fragmentation complicates routing and wastes baseline capacity. A shared index reduces duplication but requires robust metadata filtering and careful tenant testing. Benchmark both the average and worst-case filter selectivity because restrictive filters can change search behavior.

A repeatable evaluation workflow

Create a labeled set of queries with relevant document IDs and deliberately confusing near-matches. Compare embedding models, dimensions, chunking choices, distance functions, and index parameters using recall at k, mean reciprocal rank, normalized discounted cumulative gain, query latency, ingestion cost, and index size. Use exact search on a representative subset as a reference where practical.

Run evaluation again when documents, languages, embedding versions, or preprocessing rules change. Do not mix vectors from incompatible embedding models in one field. A model migration normally requires a new index or side-by-side vector field, backfill, validation, traffic cutover, and rollback plan.

The exam's best answer is rarely 'choose the largest vector.' Choose the configuration that meets measured retrieval quality, scale, latency, language, modality, security, and cost requirements, and preserve the evidence needed to repeat the decision.

Operational incident example

Assume query latency rises only for a tenant whose filter matches a very small fraction of a shared index. The foundation model is not the first suspect. Compare unfiltered and filtered vector latency, candidate counts, shard distribution, and index parameters. The remedy may be a tenant-specific index, a routing partition, different filterable metadata, or tuned search breadth. Increasing generation capacity will not repair vector search.

Monitor ingestion lag, failed writes, index size, search latency, recall canaries, hot partitions, and rejected queries. For graph indices, recall can drift after large update patterns or parameter changes even when the service is healthy. Schedule evaluation, not merely availability checks.

An embedding migration should be reversible. Create a new field or index, backfill from canonical text, run the labeled suite, shadow queries, and compare result sets. Cut over a small traffic percentage and retain the old index until quality and operations stabilize. Prevent new documents from entering only one index during the transition by dual writing or replaying a versioned change log.

For multi-index routing, define a deterministic directory from tenant, language, domain, or security class to eligible indices. Query fan-out needs a cap and a merge method. Returning the top score from incomparable embedding spaces is not valid unless scores have been calibrated.

Metadata and scale tradeoff

High-cardinality filters, broad wildcard fields, and large metadata payloads can raise cost or reduce performance. Keep searchable metadata purposeful and move verbose display data to a source store referenced by ID. Test mixed workloads: bulk ingestion can contend with search resources even when average query latency looks healthy.

For freshness, measure the time from source commit to searchable vector and the time from deletion to non-retrievability. For reliability, sample known query-document pairs continuously. Availability without recall can leave a vector service technically green while the application is functionally broken.

Capacity limits are safety controls as well as cost controls. A low maximum can reject growth; an unlimited or overly high setting can create an unexpected bill. Set limits from load evidence, alarm before them, and document what the application does when search capacity is exhausted.

Disaster recovery

Keep canonical text and metadata outside the vector index so the index can be rebuilt. Document encryption keys, network policies, schemas, embedding version, and replay order. Test restoration with retrieval canaries; a restored collection that answers health checks but lacks current embeddings is not recovered.

Cost attribution

Attribute embedding calls, vector storage, indexing capacity, query capacity, and reingestion work by environment and corpus. A dimensionality change can lower storage while increasing migration cost; an index split can improve isolation while duplicating baseline capacity. Compare total lifecycle cost with measured quality.

Delete abandoned test indices and revoke their access policies after experiments. Retaining stale corpora creates cost and a data-exposure path, especially when their metadata predates current authorization rules.

Vector migration gate

• Gate Evidence required
• Compatibility New dimension, vector type, distance function, and index mapping are supported together
• Retrieval quality Holdout queries meet recall, ranking, filter, and empty-result thresholds
• Scale Build time, storage, memory, filtered p95 latency, and concurrent query behavior pass
• Cutover Dual-write or replay closes the change gap; comparison reads show equivalent eligibility
• Rollback Prior index remains addressable until reconciliation and production acceptance finish