AI Finance Analyzer

Turn raw expense data into actionable financial insights, not just reports.

---

What This Is

AI Finance Analyzer is a developer-first AI financial decision engine. It takes raw expense data (WeChat Pay CSV, etc.) and produces structured analysis with actionable, quantified suggestions.

This is NOT another bookkeeping app, budgeting tool, or expense tracker.

This IS a programmable analysis engine that answers: "What should I do about my money?"

from analyzer import run_pipeline

report = run_pipeline("wechat_bill.csv")
print(f"Health Score: {report.analysis.health_score.total_score}/100")
# → Health Score: 72/100

for s in report.advice.suggestions:
    print(f"[{s.priority}] {s.action}")
# → [high] Reduce food spending by 20% — cut back on: 美团外卖, 饿了么
#   → Save approximately ¥300/month

How It's Different

	Traditional Tools	AI Finance Analyzer
Purpose	Record what happened	Decide what to do
Output	Charts & tables	Actionable suggestions with numbers
AI Role	Bolt-on classifier	Core decision engine
Target	End users (app)	Developers (SDK)
Data Flow	Data → Category → Chart	Data → Understanding → Behavior Analysis → Decision

Architecture

┌──────────────┐     ┌──────────────┐     ┌──────────────┐     ┌──────────────┐
│   Parser     │ ──→ │  Classifier  │ ──→ │   Analyzer   │ ──→ │   Advisor    │
│  (Input)     │     │(Understanding)│     │  (Insight)   │     │ (Decision)   │
├──────────────┤     ├──────────────┤     ├──────────────┤     ├──────────────┤
│ WeChat CSV   │     │ Rules Engine │     │ Structure    │     │ DeepSeek LLM │
│ Generic CSV  │     │ + LLM Backup │     │ Trends       │     │ Prompt Eng.  │
│ (Pluggable)  │     │              │     │ Health Score │     │ Mock Mode    │
│              │     │              │     │ Risk Alerts  │     │              │
│  ❌ No AI    │     │  ⚠️ AI Assist │     │  ❌ No AI    │     │  ✅ AI Core  │
└──────────────┘     └──────────────┘     └──────────────┘     └──────────────┘

AI is used precisely where it adds value (reasoning & language generation) and avoided where determinism matters (parsing & statistics).

Quick Start

1. Install

git clone https://github.com/yourusername/ai-finance-analyzer.git
cd ai-finance-analyzer
pip install -e ".[all]"

2. Run the Demo

python examples/quickstart.py

3. Use in Your Code

from analyzer import run_pipeline

# Full pipeline: parse → classify → analyze → advise
report = run_pipeline("your_wechat_bill.csv")

# Access structured analysis
print(report.analysis.health_score)      # HealthScore(72, grade='B')
print(report.analysis.savings_rate)       # 0.23
print(report.analysis.category_breakdown) # [CategoryBreakdown(...), ...]
print(report.analysis.risk_alerts)        # [RiskAlert(...), ...]

# Access AI-generated advice
print(report.advice.summary)
print(report.advice.suggestions)

4. Enable AI Advice (Optional)

export DEEPSEEK_API_KEY="your-key-here"

Without an API key, the system uses intelligent mock advice based on your actual data patterns.

Core Metrics

Financial Health Score (0-100)

A weighted composite score based on:

• Savings Rate (30%) — Are you saving enough?
• Essential Ratio (25%) — Is spending on necessities balanced?
• Stability (20%) — How consistent are your spending patterns?
• Diversity (15%) — Is spending spread across categories?
• Trend (10%) — Is spending going up or down?

Spending Volatility Index

Measures spending consistency using the Coefficient of Variation across monthly totals. Lower = more stable habits.

Risk Alerts

Automatic detection of: category over-concentration (>35%), low savings rate (<10%), spending spikes, and trend acceleration.

API Reference

Function	Description
run_pipeline(file)	End-to-end: file → FullReport
analyze_transactions(txs)	Analyze pre-parsed transactions
get_advice(report)	Generate advice from analysis
parse_file(file)	Parse bill file → transactions
classify(transactions)	Classify → categorized transactions

Supported Formats

• ✅ WeChat Pay (微信支付) CSV
• ✅ Generic CSV (auto-detect columns)
• 🔜 Alipay (支付宝) CSV
• 🔜 Bank statement CSV
• 🔜 Screenshot OCR

Project Structure

src/analyzer/
├── parser/          # Layer 1: Data parsing (pluggable)
├── classifier/      # Layer 2: Rules + LLM classification
├── engine/          # Layer 3: Statistical analysis
├── advisor/         # Layer 4: AI decision generation
├── models/          # Shared data schemas
├── core/            # Configuration
└── pipeline.py      # Orchestrator

Development

# Install dev dependencies
pip install -e ".[dev]"

# Run tests
pytest tests/ -v

# Run with coverage
pytest tests/ --cov=analyzer --cov-report=term-missing

# Lint
ruff check src/

Configuration

Environment variables:

Variable	Default	Description
DEEPSEEK_API_KEY	(none)	DeepSeek API key for AI features
LLM_BASE_URL	https://api.deepseek.com	LLM API endpoint
LLM_MODEL	deepseek-chat	Model name
DEBUG	false	Enable debug logging

Contributing

Contributions welcome! Key areas:

• New parser plugins (Alipay, bank formats)
• Improved classification rules
• Additional analysis metrics
• Documentation & examples

License

MIT — see LICENSE for details.