Autonomous CFO: AI 3-Way Match

• 

  Autonomous CFO Dashboard: Real-time 3-Way Matching metrics & HITL Exception Queue handling invoice discrepancies.

• 

  Verified General Ledger (GL) Postings with automated account coding and payment trigger status.

• 

  Agent Orchestrator (AO) task management board showing verified agent executions & passing unit tests.

• 

  vs code orchesteror.py

• 

  Audit-compliant Neatlogs trail tracking AI reasoning, human approvals, and GL posting events.

• 

  execute the task in the agent orchestrator

Inspiration

Accounts Payable (AP) teams spend hundreds of hours manually comparing Purchase Orders (POs), Receiving Logs (GRNs), and Vendor Invoices. Manual 3-Way Matching leads to human oversight, delayed supplier payments, undetected quantity shortages, and incorrect General Ledger (GL) postings. We built Autonomous CFO to automate the complex matching process, detect price and quantity variances instantly, and route only genuine exceptions to human controllers.

What it does

Autonomous CFO is an AI-powered financial automation engine that executes deep 3-Way Matching across enterprise financial documents:

• Automated 3-Way Matching: Compares POs, GRNs, and Invoices using LLM intelligence via TensorMux API.
• Smart Exception Taxonomy: Identifies specific issues like QUANTITY_SHORTAGE, PRICE_VARIANCE, UNAPPROVED_FREIGHT, and BANK_DETAIL_MISMATCH.
• Human-in-the-Loop (HITL) Dashboard: Allows human controllers to review flagged exceptions, approve override postings, or send dispute notifications directly to vendors.
• Deterministic Fallback Engine: Features a math rule engine to calculate financial discrepancies even if API connections fluctuate.
• Audit Trail Compliance: Uses Neatlogs for audit-compliant event logging, GL entry postings, and compliance report exports.

How we built it

• Frontend Dashboard: Built with Streamlit and customized with dark-mode CSS styling and Plotly charts for real-time visual executive analytics.
• AI Intelligence & LLM: Integrated TensorMux API using glm-4-7-flash model via the OpenAI SDK for structured audit workpaper generation and context evaluation.
• Orchestration: Built a custom AgentOrchestrator to maintain state machines across document batches, batch summaries, and HITL queues.
• Logging & Audit Compliance: Integrated Neatlogs for tracking system lifecycle events, approval workflows, and generating compliance exports (JSON/CSV).

Challenges we faced

• Accurate Discrepancy Math: Ensuring that quantity shortages were multiplied against unit prices rather than relying solely on price variances required refining our fallback engine logic.
• Structured LLM Outputs: Crafting strict system prompts to guarantee valid JSON structures containing GL account codes, risk severity, and audit notes.

Accomplishments that we're proud of

• Achieving end-to-end processing with automated GL journal entry generation for matching invoices.
• Creating an executive-ready dark CFO dashboard with Plotly charts and glassmorphism design.
• Implementing a complete Human-in-the-Loop workflow that safely bridges AI recommendations with human oversight.

What we learned

Building an autonomous financial agent requires balancing flexible LLM reasoning with deterministic accounting rules. Reliable AP automation relies heavily on state management, audit trails, and graceful fallbacks.

What's next for Autonomous CFO

• Direct API connectors for enterprise ERPs (SAP, NetSuite, QuickBooks).
• Automated vendor email outreach for disputed invoices.
• Multi-currency conversion and foreign exchange variance handling.

Built With

• json
• neatlogs
• openai
• plotly
• python
• streamlit
• tensormux