Financial Close & GL Variance Investigation - Projects

The business problem

A controller asks, “Why did operating expenses rise 18% this month?” Answering it means comparing periods, isolating abnormal accounts and drilling from GL balances to journals and subledger transactions.

Proposed solution

The crew decides what to investigate. It compares periods, isolates abnormal accounts, and drills from GL balances to journals and subledger transactions, correlating cost centers and suppliers.

It answers with evidence, for example: “Marketing rose $286K; about 74% came from three journals for the annual customer conference; no unusual manual-journal activity.” It then drafts flux commentary and proposed adjusting journals for review.

Workflow

Collect General-Ledger Balances
Compare Against Prior Period and Budget
Detect Significant Variances
Drill Down to Journals and Subledgers
Investigate Root Cause and Reconcile
Prepare Draft Adjustments for Review
Obtain Controller Review
Record Audit Trail

Agents and responsibilities

Variance Detection: actual vs. prior period, budget and forecast against materiality thresholds.
Drill-down: GL to journals to subledger: AP, AR, FA, payroll.
Explanation: RAG over accounting policies, the close checklist and prior commentary.
Reconciliation: intercompany and suspense accounts.
Journal Proposal: draft adjusting entries only, never posted by an agent.

Oracle ERP integration

Oracle objects: Ledgers, period balances, journal batches, headers and lines, chart-of-accounts segments and subledger accounting entries; E-Business Suite GL_BALANCES and GL_JE_HEADERS/LINES.

Shared Oracle integration: a Python MCP server exposing typed read tools (Fusion Cloud REST resources, BI Publisher/OTBI reports, read-only SQL views for E-Business Suite) and a small set of write tools that execute only after human approval.

Technology stack

Project stack: Claude Agent SDK (subagents for drill-down) · CrewAI + MCP · Gemini and GPT runtimes · read-only SQL and OTBI tools

Orchestration: CrewAI crews define role-based agents, tasks and hand-offs. Oracle tools attach through CrewAI MCP server adapters, so every agent calls the same governed tool layer.
Agent runtimes: Claude Agent SDK with Claude is the primary build and reasoning runtime. It uses subagents for deep investigation, MCP for tool access, and pre-tool-use hooks that block any ERP write lacking an approval token. The same agents run on Google ADK with Gemini and OpenAI Agents SDK with GPT through a common model-adapter interface, switched by configuration.
Engineering: Python, Pydantic-validated agent outputs, a pytest scenario-evaluation harness comparing Claude, Gemini and GPT on accuracy, latency and cost, OpenTelemetry tracing and Docker, built in VS Code with Claude Code.

Controls and safeguards

SQL runs only against approved read-only views with row limits.
All figures are computed by SQL, never by the model, and each explanation cites its journals.
Proposed journals pass through standard GL approval.
Grounding and control (shared platform): RAG over policies, contracts and approval matrices in Oracle Database 23ai AI Vector Search; a deterministic rule engine for tolerances, thresholds and SoD; an immutable audit log of evidence, tool calls, model, approver and outcome.

Validation plan

Scope: A ledger seeded with accrual misses, misclassifications, FX movements and intercompany mismatches.

Targets

Reference build in progress. The figures below are validation targets for the build, not measured results.

95%+ Seeded variances found
85%+ Correct root cause