In many construction companies across developing economies, traditional paper-based accounting practices remain dominant. Invoices, receipts, and material logs are often handwritten, scanned, or photographed, making digital processing and compliance reporting inefficient, error-prone, and labor-intensive. The Cognaptus AI Accounting Assistant aims to solve this problem by delivering a modular, user-friendly, AI-enhanced accounting system that can ingest these documents and transform them into structured digital workflows—without requiring technical skills from the end user.

This demo showcases the user interface of our solution, which supports seamless access via both desktop web browsers and mobile applications. Users can scan receipts using their phone cameras, approve transactions on the go, and view reports and dashboards in real time across devices. This mobility ensures that site engineers, finance officers, and senior managers remain synchronized with minimal operational friction.

Overview of the Solution Interface

The solution is organized into intuitive stages, simulating a real accounting workflow:

1. Upload & OCR Extraction

  • Upload scanned receipts/invoices using mobile or desktop
  • Extracts:
    • TIN number
    • Invoice type and number
    • Material description
    • Quantity, price, total, and date
  • Editable fields after extraction

Current Model: TrOCR (Transformer-based OCR)

Future Plan: Fine-tune on Filipino invoice layouts and handwritten forms

2. Approval Routing

  • Suggests approver based on amount and nature
  • Users see live approval status
  • Proceeds to journal entry after approval

Current: Rule-based logic

Future: Fine-tuned classification model (T5/BERT)

3. Suggested Journal Entry

  • Add sub-account tags (department, project, activity)
  • System generates editable journal entry + explanation

Current: Template-based logic

Future: Fine-tuned encoder-decoder model (e.g., T5)

4. Tax Declaration Page

  • Auto-fills BIR Forms:
    • 2307 (withholding)
    • 2550M (VAT)
    • 1601-EQ (quarterly withholding)
  • Direct integration with E-BIR system for submission

Future Plan: GPT-style model for tax recommendation/validation

5. Summary Reports

  • Shows total expenses, VAT, vendor stats
  • Includes:
    • Activity-based cost analysis
    • Historical cost benchmarking
    • Macroeconomic comparison (e.g., fuel, cement trends)

Future: LLM + news scraping for annotations

6. Construction Project Monitoring

  • Shows:
    • Project progress
    • Gross profit (Revenue - Cost)
    • Cost breakdown (labor, material, logistics, overhead)
    • Anomaly detection example: “Labor over 22% for Riverpark”

Planned Model: LSTM/statistical anomaly detection + LLM reporting

7. Project Bidding & Marketing Allocation

  • Shows:
    • Bid status per project
    • PR/marketing expense tied to each bid
    • Predictive success model

Future Model: Classifier trained on bidding outcomes

8. Cash Flow Dashboard

  • Historical inflows/outflows
  • Forecasts next 6 months
  • Generates financing alerts (e.g., reserve shortfall)
  • Embedded with macroeconomic signals (e.g., inflation risk)

Planned Model: Prophet + LLM overlay

System Architecture & AI Deployment

  • Local server runs all AI models
  • Secure web/mobile access for users
  • Scalable, private, and responsive

Manager-Friendly, No-Code Adaptability

Managers can adapt the system using natural language. Examples:

  • “Add rule: Expenses >₱100,000 require CFO approval”
  • “Group marketing under BID-2025-Mall”
  • “Create cost/m² report for all projects”

Fine-tuned LLM acts as a semantic rules engine to parse and apply instructions

Final Thoughts

This interface is a glimpse into Cognaptus’ broader vision: digitizing and automating accounting across paper-heavy industries—intuitively, flexibly, and intelligently, without needing developers.

Accessible. Local. Smart. Cognaptus.