Published Paper Details:

Automated personal finance management intersects two distinct but deeply interrelated challenges: precise transaction-level semantic categorization and jurisdiction-specific statutory tax compliance. In India, these challenges are compounded by a dual-regime income tax structure introduced under the Finance Act 2020 and substantially reformed in 2023, requiring taxpayers to compare liabilities under both regimes before filing. This paper presents FinanceAI, a full-stack intelligent system that unifies personal finance management with an end-to-end Indian income tax engine. The core technical contribution is a confidence-gated hybrid classification pipeline that applies deterministic keyword-driven rules for high-confidence transactions while delegating ambiguous items to an LLM (LLaMA-3.1-8B-Instant via Groq) at a configurable confidence threshold (? = 0.70). The system implements complete FY 2024-25 tax slab logic for both the Old and New regimes, including Chapter VI-A deductions, rebate under Section 87A, and standard deduction differentials. Classified transaction streams are automatically converted into double-entry accounting ledger and journal entries. Downstream, a document generation engine produces portal-compatible ITR-1 Sahaj JSON, Form 16 Tax Computation Certificates, and Form 26AS TDS Reconciliation PDFs using ReportLab. A context-aware AI Chartered Accountant (AI-CA) chatbot grounded in the user's live financial data provides personalized advisory backed by an offline rule fallback. Experimental analysis demonstrates that the hybrid pipeline achieves substantially higher classification accuracy on domain-specific Indian financial transactions compared to a purely rule-based baseline, while maintaining deterministic fallback integrity. The system operates fully locally with no mandatory external API dependency, positioning it as a viable open-source alternative to proprietary tax-filing platforms. Results confirm the feasibility of LLM-augmented statutory compliance in resource-constrained personal finance contexts.