Research Oracle Fusion Cloud APIs related to testing. Identify key endpoints across modules (e.g., Finance, HCM), and document how they can be used for automating test scenarios like functional, integration, and data validation testing. Summarize findings in a document and provide a reference table.

A task assigned to the team (notably Divyanshu Jaiswal) to collect and organize official Oracle Fusion Financials documentation in PDF format. The objective was to create an offline, structured knowledge repository from https://docs.oracle.com. The process included navigating all “Top Task” links, downloading PDFs, structuring files with consistent naming, and packaging them into a ZIP archive for internal use. The archive will support downstream processing like embedding and ingestion.

Phase 1 focuses on laying the groundwork for the ERP AI agent by developing a semantic, vectorized knowledge base. The deliverables include building a high-fidelity semantic index using transformer-based embeddings, metadata-rich taxonomy schemas, document ingestion pipelines, and governance protocols. This phase emphasizes extensibility, auditability, and retrieval quality, setting the stage for NLP, tool invocation, and future interaction layers in the AI agent.

This document details a cost-efficient deployment strategy for the AI Agent using Oracle Cloud Infrastructure (OCI) and AWS Spot GPUs. It includes tenancy planning, compartment-based access control, landing zone setup, and provisioning of compute environments. A hybrid infrastructure using OCI’s Always Free services and AWS g4dn Spot Instances ensures under-$20 monthly GPU costs for RAG tasks. This infrastructure is designed to support both inference and training workloads for the ERP agent.

This foundational white paper introduces the vision, architecture, and roadmap for building an AI Agent tailored for ERP system implementation. It outlines the modular architecture including a semantic knowledge base, NLP understanding, task automation, machine learning insights, and API integration. The agent is designed to reduce implementation costs, improve consistency, and enable intelligent automation across ERP modules like Oracle Fusion. Phased deployment strategy from PoC to specialization is detailed along with cost and future expansion plans.

This task involved designing and implementing a full Retrieval-Augmented Generation (RAG) pipeline using ERP PDF documentation. It includes preprocessing documents, generating semantic embeddings using all-MiniLM-L6-v2, storing them in Qdrant, and enabling semantic search through a CLI. The system supports structured metadata, fast retrieval using cosine similarity, and is designed to scale into a full RAG-powered ERP assistant. Future enhancements include LLM-based synthesis and UI integration.