[Janaka2]

Motivation

• Introduce practical, testable AI application patterns for prompt assembly, lightweight RAG, model orchestration, tool integration, safety/policy enforcement, and repeatable evaluation.
• Provide small, composable building blocks that teams can copy into services and unit-test locally.

Description

• Add new package org.example.ai containing PromptBuilder, RetrievalPipeline, LlmClient (interface), FallbackModelStrategy, ToolAgentFacade, PolicyEnforcedLlmClient, and EvaluationHarness to implement the patterns above in a focused, library-style API.
• PromptBuilder implements template substitution, and RetrievalPipeline separates retrieval and reranking and produces a bounded context window for prompts.
• ToolAgentFacade routes tool:<name>:<input> tool invocations and composes retrieval context into LLM prompts via LlmClient, while FallbackModelStrategy fails over across multiple LlmClient implementations.
• PolicyEnforcedLlmClient decorates an LlmClient with input validation, PII masking, and output filtering policies, and EvaluationHarness provides golden-set and statistical evaluation helpers; README.md and docs/phase-5-ai-era-patterns.md were updated to document the phase and examples.

Testing

• Added AiApplicationPatternsTest which asserts RAG-lite + facade orchestration, tool invocation, fallback model behavior, policy masking/filtering, and golden/statistical evaluation flows.
• Ran mvn -Dtest=AiApplicationPatternsTest test in this environment but the build failed to start tests due to external artifact resolution failure (Maven Central returned HTTP 403 resolving maven-enforcer-plugin:3.5.0).
• Because of the Maven resolution error the new tests were not executed here, but they are included in the repository and should run in a network-enabled CI environment where Maven Central is accessible.

---

Codex Task