AI Architecture Patterns
Overview
Overview
Responsible AI practices, compliance frameworks, data protection, and ethical considerations for AI systems.
How to implement AI tool calling (function calling) in Salesforce Apex, working around the DML-before-callout transaction limitation with a multi-transaction follow-up pattern.
A complete guide to building a production-quality AI chatbot with the Anthropic Claude API, covering architecture, streaming responses, mobile-first UI, and the CSS patterns that make chat widgets work reliably on every device.
Strategies for managing multi-turn LLM conversation token costs by compacting older messages into summaries while preserving context continuity.
Using Custom Metadata Types to externalize AI chatbot configuration -- model settings, behavior rules, and UI prompts -- with cached Apex queries and fallback defaults.
How to architect multi-step conversational workflows through system prompt engineering and Custom Metadata, keeping flow logic declarative and deployable without code changes.
Salesforce-native AI capabilities — Data Cloud, Einstein, Agentforce, Prompt Builder, and the Trust Layer.
How and why to replace open-ended text input with structured, selectable choices in AI chat interfaces — covering the design principle, a real-world Claude Code example, and implementation patterns across React, WhatsApp, Slack, Telegram, Messenger, and Microsoft Teams.
How to prevent INVALID_OR_NULL_FOR_RESTRICTED_PICKLIST errors when AI tools create Salesforce records, by dynamically populating tool enum values from org metadata.