In many construction firms across developing economies, accounting still revolves around handwritten or scanned documents—invoices, receipts, and material logs. These paper-based workflows make reporting slow, error-prone, and expensive.

Cognaptus AI Accounting Assistant addresses this gap by offering a modular, intuitive interface that transforms messy documents into structured, AI-augmented workflows. No technical background needed.

🖥️ Demo Overview

This demo simulates a full accounting flow via a web/mobile interface:

  • Upload receipts from your phone
  • Extract and validate key data
  • Route approvals
  • Auto-generate journal entries, tax forms, and reports
  • Forecast cash flow and flag risks

Cross-device compatibility ensures that site engineers, finance staff, and executives stay aligned with minimal friction.

1. Upload & OCR Extraction

  • Upload via mobile or desktop
  • Extracts:
    • TIN number
    • Invoice type, number
    • Description, quantity, amount
    • Date of transaction
  • Fields are editable post-scan

OCR Extraction
Editable Fields

Current Model: TrOCR (Transformer OCR)
Planned: Fine-tune on Filipino receipts and handwriting

2. Approval Routing

  • Suggests approver based on amount/type
  • Displays live status
  • Pushes approved entries to journal draft

Approval Routing

Current: Rule-based logic
Planned: Classification model (T5 or BERT)

3. Journal Entry Suggestion

  • Add department, project, or activity tags
  • Generates editable journal entry + explanation

Journal Suggestion

Current: Template logic
Planned: Encoder-decoder model (e.g. fine-tuned T5)

4. Tax Form Auto-Fill

  • Supports:
    • BIR 2307 (withholding)
    • 2550M (VAT monthly)
    • 1601-EQ (quarterly withholding)
  • Plans for E-BIR direct integration

Tax Form Automation

Planned: GPT-style validation & recommendation engine

5. Reports & Cost Analytics

  • Tracks:
    • Total expenses & VAT
    • Vendor spend breakdown
    • Activity-based costing
    • Cost benchmarking (e.g., fuel, cement)

Reports

Planned: LLM + news scraping for dynamic annotations

6. Project Cost Monitoring

  • View:
    • Project profit (revenue – cost)
    • Cost breakdown (labor, material, logistics)
    • Anomalies (e.g., labor over 22% on Riverpark project)

Project Monitoring

Planned: LSTM-based anomaly detection + LLM report assistant

7. Bid & Marketing Allocation

  • Monitor:
    • Bid status per project
    • Marketing costs tied to bids
    • Success prediction model (coming soon)

Bid Tracking

Planned: Outcome classifier trained on historical bids

8. Cash Flow Forecasting

  • Historical cash trends
  • Forecasts next 6 months
  • Flags financing gaps (e.g., reserve shortfall)
  • Tracks macroeconomic signals (e.g., inflation)

Cash Flow

Planned: Prophet + LLM overlay for smart alerts

🧱 System Architecture

  • Hosted on local secure servers
  • Accessible via browser or mobile app
  • Scalable, offline-ready, and private by design

🧠 Natural Language Customization

Managers can use plain English to set rules and reports:

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

Fine-tuned LLM interprets and applies these commands using a semantic rules engine

🌟 Final Thoughts

This prototype illustrates how AI can bridge analog operations with digital automation—especially in construction industries where paperwork still dominates.

Cognaptus is building tools that are:

Accessible. Private. AI-native. Designed for the real world.