Integrated Human-in-the-Loop (HITL) workflow controlled via the Transformation Control desk ensures AI-powered data transformations governance, integrity, and transparency.
AI
integration
framework
Intelligent commerce transformation implies extensive AI integration for both personalizing customer experience and streamlining operational efficiency to maximize business outcome.
While Salesforce B2C Commerce (SFCC B2C) platform offers truly outstanding customer engagement capabilities with Einstein, the challenge of AI-powered merchandising and operations naturally requires a business-specific solution.
Challenge accepted. AI Transformer is the industry-first AI Operations (AIOps) framework for Salesforce B2C Commerce.
Intelligent Commerce Transformation Challenges
E-commerce industry's movement toward an AI-driven future, especially in merchandising and operations, introduces multiple challenges, including defining the transformation roadmap, designing a robust AI integration architecture, maintaining brand safety and governance compliance, and ensuring cost-predictable AI service utilization.
AI Transformer is designed to efficiently address these challenges, and it is based on the following core pillars to achieve that.
TRUSTED AI GOVERNANCE &
BRAND SAFETY
ACCELERATED TIME-TO-VALUE &
MERCHANDISING VELOCITY
Comprehensive Transformation Playground console provides in-place access to crucial framework integration endpoints for rapid prototyping and implementation support.
SEAMLESS AI INTEGRATION &
FUTURE-PROOF ARCHITECTURE
Hooks-based integration design, alongside framework-managed Reusable Prompt and Data Model Registries, ensures seamless AI enablement and high architectural scalability.
AI Transformer is a platform-hosted AI Operations framework for Salesforce B2C Commerce, designed to establish a future-proof AI integration foundation and maintain a governance-first commerce data transformation workflow, leveraging OpenAI Platform capabilities.
Transformation Control
AI Transformer delivers Transformation Control desk for Business Manager, the command center interface for trusted AI governance and brand safety. It maintains dedicated oversight for data transformations and ensures full control over critical Human-in-the-Loop workflow.
Control Engine
Integrated control engine is built on the following core technological concepts defining its operating principles and working methodology:
- Transformation Records: Control engine implements caching all transforming data, prompt references, transformation results, statuses and other details to the platform-stored records.
- Transformation Status: Ongoing transformation state tracking implements and maintains the transformation lifecycle.
- Workflow Automation: Transformation results acceptance and appliance is configurable to work in one of strict, human-controlled mode, or oversight, fully-automated mode.
- Decision Tracking: Governance actions are immutably logged with User ID, timestamp, and applied action type, ensuring full accountability for changes across the entire workflow.
- Workflow Protection: Automated verification of transforming data, transformation results, and applied governance actions maintained across the transformation lifecycle.
Transformation Playground
AI Transformer offers a multi-featured Transformation Playground console directly within Business Manager enabling AI interaction through the framework-delivered integration endpoints. It streamlines prompts prototyping, data modeling, and AI processing verification.
Construction Kit
These functions de-facto define the data transformation workflow, starting from prototyping and ending with invoking bound Reusable Prompts defined at OpenAI Platform side for data processing.
- Reference Prompting: Showcases the AI's stateless behavior when invoked programmatically for common operations.
- Data Modeling: Enables rapid prototyping of free-form data models and parametrized prompts for common operations.
- Data Normalization: Provides seamless conversion of diverse structured text data to JSON-form for parametrized prompting.
- Model Prompting: Enables parametrized prompts constructions and verification against the instance's pre-defined data models.
- Reusable Prompting: Facilitates calling bound OpenAI Reusable Prompts with the instance's real data for transformation insights, quality assurance and optimization purposes.
Specialized Utilities
These functions provide convenient oversight and verification of the integration endpoints functioning and general AI behavior.
- Text & Image Moderation: Demonstrates the AI's capabilities for content moderation, available through the framework.
- Text Translation: Showcases framework-established translation engine capabilities for high-quality multilingual translation.
- Image Analysis: Demonstrates the AI's capabilities for analyzing images by public URL with flexible prompting supported.
- Image Generation: Showcases the AI mechanism by displaying both the generated image and the prompt AI refined for applying.
Integration Framework
AI Transformer is organized into two SFCC B2C cartridges defining its integration specifics and delivered functionality: int_ai_transformer core integration cartridge (framework engine, OpenAI Platform integration and reusable functional endpoints); and bm_ai_transformer Business Manager Extension cartridge (construction tools and HITL control desk).
Integration Design
AI Transformer leverages SFCC B2C custom hooks to ensure a weak-dependency integration of its functional endpoints.
This approach provides the following critical advantages:
- Minimal Effort: As no SDK installation or fine-tuning is required, integration is reduced to hooks injection and data preparation.
- Reliability: Hooks natively ensure decoupling of AI-powered integration functions and core business logic.
- Compatibility: Hooks supported by all reference architectures, including the latest PWA Kit, SFRA, and even vanilla SiteGenesis.
Integration Endpoints
AI Transformer delivers a range of specialized functional endpoints, utilizing embedded integration with crucial OpenAI Platform APIs.
These integration endpoints provide the following core functionality:
- Text & Image Moderation: Framework-orchestrated access to the dedicated /v1/moderations API ensures performant, cost-efficient, and secure text and image moderation.
- Data Processing: Framework-provisioned data processing engine combines the /v1/responses API and Reusable Prompting for maximum flexibility and control in data operations.
- Text Translation: Framework-established translation engine utilizes the /v1/chat/completions API to deliver language detection, auto-translation, and target translation functions, while supporting translation context and creativity control.
- Image Analysis: Managed access to /v1/chat/completions API ensures configurable and cost-efficient image analysis.
- Raw Calling: Direct invoking of the de-facto industry-standard /v1/chat/completions API enables advanced use cases handling.
Data Processing Foundation
Reusable Prompts is a recent, enterprise-grade OpenAI Platform addition, optimizing, securing, and structuring data processing.
AI Transformer managed Reusable Prompt bindings and structured Data Model configurations directly reflect this approach, delivering the following strategic advantages:
- Integrity: Code-level Registries ensure persistent Data-to-Prompt mapping to maintain data processing integrity.
- Consistency: Fixed Reusable Prompts and Data Models enforce predictable behavior and consistent, use case agnostic output.
- Scalability: Leveraging once-defined Data Models and Reusable Prompts across multiple scenarios optimizes scaling efforts.
- Optimization: Leveraging the Reusable Prompts eliminates text prompt transmission and reduces data volume.
- Efficiency: Structured data processing minimizes token usage.
- Maintenance: Prompt Template updates are automatically propagated to all relying AI-powered functional endpoints.
Immediate Value Deliverables
Designed as a solid, generic framework, AI Transformer enables seamless addressing of the full spectrum of SFCC B2C back-end use cases.
The Jobs provisioned as a part of the AI Transformer bundle showcase practical handling of one of the most frequent and crucial intelligent commerce requests, Products and Categories SEO Maintenance with AI, immediatelly delivering the following values.
Maintain Product Meta Descriptions Job
Generates meta descriptions for new products or when missing using OpenAI Platform and manages approved changes appliance.
- Controlled Processing: Configurable product last update / create time filter controls computing resource consumption by defining a precise subset of products eligible for processing.
- Controlled AI Utilization: Service calls limit setting guarantees predictable OpenAI Platform utilization, directly managing operational costs.
- AI Failures Tracking & Mitigation: Process-level circuit breaker configuration enables controlled termination during AI service unavailability, mitigating hung risks and ensuring system stability.
- Localized Processing: Locale setting ensures the entire process is localized, from data construction to results appliance.
- Appliance Automation: Auto-approve and appliance threshold day settings maintain both strict control (human verification) and oversight (automated) transformation modes.
- Adaptivity & Configuration: Job's source code full access allows for adaptation to any specific business request, while Prompt Templates are fully customizable to tune AI behavior.
Maintain Category Meta Descriptions Job
Updates category meta descriptions on a periodic basis taking into account categories hierarchy changes if any, leveraging OpenAI Platform and manages approved updates appliance.
The Job follows an identical design pattern as the one maintaining product meta descriptions and delivers all the same crucial valuable advantages: processing control, AI utilization control, process localization, appliance automation, and adaptation flexibility.
The main functional difference between the two Jobs lies in the SEO persistency requirement handling: product meta descriptions prioritize stability, while category meta descriptions must actively address the dynamic nature of catalog structure changes.
What's Next?
First of all, we would like to thank you for your interest in AI Transformer and for reading this detailed, but hopefully insightful overview.
There are many more technical details and specialized use cases we wanted to showcase, but this overview could become a book 😆.
Answering the question of what's next: more exciting features to AI Transformer, and more AI-powered solutions from dwreTeam.
Contact us to discuss how AI Transformer can help address your specific business needs. We are ready to align it for you.