AI Accountant

AI Accountant is a software-only AIML application that converts raw transaction data into structured double-entry accounting records and professional financial statements.

Core Input

The initial supported input format is:

Date Transaction Amount

---

2026-04-01 Started business with cash 100000 2026-04-02 Purchased goods from Ram on credit 25000

Core Workflow

Excel / CSV
    ↓
File Validation & Normalization
    ↓
AI Transaction Understanding
    ↓
Confidence Scoring
    ↓
Human Review (when required)
    ↓
Validated Accounting Transaction
    ↓
Accounting Engine
    ↓
Journal
    ↓
Ledger
    ↓
Trial Balance
    ↓
Trading A/c
    ↓
P&L A/c
    ↓
Balance Sheet
    ↓
AI Anomaly Detection & Financial Insights

Main Objectives

1. Understand natural-language accounting transactions.
2. Convert them into valid debit/credit entries.
3. Generate Journal, Ledger and Trial Balance automatically.
4. Generate professional Trading Account, P&L Account and Balance Sheet.
5. Detect ambiguous transactions and request human verification.
6. Detect unusual accounting behaviour using ML.
7. Support large transaction files through background processing.
8. Maintain complete traceability of AI decisions and user corrections.

Architecture Principle

AI interprets the transaction. The deterministic accounting engine applies accounting rules.

The AI must never directly calculate financial statements. All accounting outputs must pass deterministic validation.

Technology Stack

Frontend

• Next.js
• TypeScript
• Tailwind CSS
• Responsive desktop/tablet/mobile UI

Backend

• FastAPI
• Python
• Pydantic
• SQLAlchemy
• Alembic

AIML

• pandas / NumPy
• scikit-learn
• LightGBM / XGBoost
• Sentence Transformers
• spaCy
• Optional PyTorch-based models where justified

Data

• PostgreSQL
• pgvector
• Redis
• Celery

DevOps

• Docker
• GitHub Actions
• Production deployment

Core AIML Modules

1. Transaction Understanding

Transaction Text
    ↓
Preprocessing
    ↓
Entity Extraction
    ↓
Transaction Classification
    ↓
Account Identification
    ↓
Debit/Credit Prediction
    ↓
Confidence Score

2. Semantic Matching

Used to find similar historical transactions and improve consistency.

3. Anomaly Detection

Uses historical accounting behaviour to identify unusual amounts, account combinations and transaction patterns.

4. Human-in-the-Loop Learning

AI Prediction
    ↓
User Correction
    ↓
Stored Correction
    ↓
Future Training Data

Accounting Validation Rules

Every journal entry must satisfy:

Total Debit = Total Credit

The financial statements must satisfy:

Assets = Liabilities + Capital

The system must fail safely instead of presenting an invalid Balance Sheet.

Frontend Routes

/login
/dashboard
/upload
/ai-review
/accounting/journal
/accounting/ledger
/accounting/trial-balance
/financial-statements
/insights
/settings

Documentation

• PROJECT_ARCHITECTURE.md --- complete system architecture and phase flow
• BACKEND_ARCHITECTURE.md --- backend services, database interaction, APIs and frontend integration
• FRONTEND_ARCHITECTURE.md --- pages, components, responsive behaviour and frontend state/data flow
• DATABASE_INFO.md --- database schema, tables, relationships, indexes and accounting data model

Development Strategy

Development is phase-driven. Each phase must be implemented, tested and validated before dependent work is considered complete.

Foundation
    ↓
Database Model
    ↓
Accounting Engine
    ↓
Dataset Engine
    ↓
AI
    ↓
AI + Accounting Integration
    ↓
Human Review
    ↓
Anomaly Detection
    ↓
Backend APIs
    ↓
Frontend
    ↓
Export
    ↓
Security
    ↓
Performance
    ↓
Testing
    ↓
Deployment

Important Scope Rule

This project is software-only. No hardware is required.

Project Status

Initial architecture and implementation specification.