DEV Community

兆鹏 于
兆鹏 于

Posted on

104 Financial AI Skills, Zero API Cost: Pure Python, Millisecond Response

Zero-API Financial AI Skills Library: 104 Scenarios in Pure Python, Millisecond Response

The API Bill Problem in Financial AI

Here's a paradox I kept running into in financial AI: business teams want "instant and ready," but tech teams deliver "integrate the API first."

One invoice verification scenario — hook into a third-party OCR service, 0.3 RMB per call, 100K calls a day, that's 90K RMB/month. One financial report analysis need — call an LLM API, 0.8 RMB per request in token costs, batch process 500 reports during earnings season, burn 400 RMB in a day. And don't get me started on budget control and risk scoring — those high-frequency scenarios where API costs scale linearly with volume. By year-end reconciliation, your AI project's ROI might be worse than a traditional Excel macro.

This isn't a capability problem. It's an architecture decision problem.

The financial-ai-skills repo takes a different path: pure Python standard library, zero API cost, millisecond response, 104 financial scenarios ready to go.

Repo: https://github.com/yuzhaopeng-up/financial-ai-skills


1. The Full Picture: 104 Skills Architecture

financial-ai-skills isn't a demo project — it's an engineered financial AI skills library covering 23 major categories and 104 specific Skills, published as 7 standalone Python packages.

1.1 Layered Scenario Architecture

┌─────────────────────────────────────────────────────────┐
│                  financial-ai-skills                     │
│                   104 Skills / 23 类                     │
├──────────────┬──────────────┬────────────────────────────┤
│  Financial   │   Wealth     │   Risk & Compliance        │
│  Intelligence│   Management │                             │
│  6大类       │   8大类      │   9大类                     │
│  27 Skill    │  36 Skill   │   41 Skill                  │
├──────────────┴──────────────┴────────────────────────────┤
│            Infrastructure Skills (通用层)                 │
│   wecom-template-card │ customer-marketing               │
│   product-manual-rag  │ application-material-checker     │
└─────────────────────────────────────────────────────────┘
Enter fullscreen mode Exit fullscreen mode

1.2 Seven Published Packages

Package Domain Skills Core Capabilities
financial-intelligence Financial Intelligence 27 Invoice verification, budget control, report analysis, tax calculation, cost accounting, fund forecasting
wealth-management Wealth Management 36 Asset allocation, portfolio analysis, product recommendation, risk assessment, portfolio optimization
risk-compliance Risk & Compliance 41 Enterprise risk assessment, AML detection, compliance review, credit scoring, alert monitoring
wecom-template-card IM Output - Markdown to WeCom/Feishu/DingTalk template cards, one-click adaptation for all three IM platforms
customer-marketing Customer Marketing - Customer profiling, precision marketing, churn prediction, campaign effectiveness
product-manual-rag Product Knowledge - Product manual RAG retrieval, clause parsing, comparison recommendations
application-material-checker Material Review - Account opening review, loan application check, compliance document verification

1.3 Key Design Principles

Zero external dependencies: All Skills rely solely on the Python standard library (json, re, datetime, math, decimal, etc.) — no third-party APIs or LLM services. This means:

  • Install and use immediately, no API key needed
  • Response times in milliseconds (no network I/O)
  • Cost is always zero, regardless of call volume
  • Deployable in intranet / offline environments

Built-in mock data: Each Skill package ships with complete mock datasets — experience full functionality without configuring a database. Swap in real data sources for production.


2. Hands-On: 5 Typical Scenarios

2.1 Invoice Verification — Instantly Check Authenticity

$ python financial_cli.py invoice 011001900111 12345678
Enter fullscreen mode Exit fullscreen mode
Field Value
Invoice Code 011001900111
Invoice Number 12345678
Verification Result Invoice information matches
Issue Date 2025-03-15
Buyer Name Beijing Technology Co., Ltd.
Amount 12,800.00
Tax 1,664.00
Total (incl. tax) 14,464.00
Status Valid

Invoice verification is a high-frequency operation in financial shared service centers. The traditional approach hooks into the tax bureau API or a third-party OCR — dealing with network timeouts, rate limiting, and billing reconciliation. The financial-intelligence package does it differently: local rule engine + validation algorithms, entire process under 5ms.

Core code:

from financial_intelligence import InvoiceChecker

checker = InvoiceChecker()
result = checker.verify("011001900111", "12345678")
print(result.to_markdown())
Enter fullscreen mode Exit fullscreen mode

2.2 Budget Control — Instant Overspend Alerts

$ python financial_cli.py budget 市场部
Enter fullscreen mode Exit fullscreen mode
Category Budget Used Utilization Status
Ad Spend 500,000 523,400 104.7% Over Budget
Event Execution 300,000 287,600 95.9% Warning
Media Partnerships 200,000 178,300 89.2% Normal
Brand Building 150,000 156,200 104.1% Over Budget
Total 1,150,000 1,145,500 99.6% Critical

The pain point with budget control isn't "can't calculate" — it's not fast enough, not pushed in time. Millisecond response from local Skills makes real-time budget gatekeeping possible.

from financial_intelligence import BudgetEngine

engine = BudgetEngine()
report = engine.check_department("市场部")
print(report.to_markdown())
Enter fullscreen mode Exit fullscreen mode

2.3 Financial Report Quick Read — Key Metrics & YoY at a Glance

$ python financial_cli.py report 美的集团 2025
Enter fullscreen mode Exit fullscreen mode
Metric 2025 2024 YoY Change Trend
Revenue 4,023B 3,737B +7.7% Up
Net Profit (attributable) 385B 337B +14.2% Strong Up
Gross Margin 26.8% 25.3% +1.5pp Up
Net Margin 9.6% 9.0% +0.6pp Up
ROE 24.3% 22.8% +1.5pp Up

The report quick-read Skill supports metric extraction, YoY/QoQ calculation, trend judgment, and core conclusion generation.

from financial_intelligence import ReportReader

reader = ReportReader()
summary = reader.analyze("美的集团", year=2025)
print(summary.to_markdown())
Enter fullscreen mode Exit fullscreen mode

2.4 Asset Allocation — Personal Wealth Management Engine

from wealth_management import WealthEngine

engine = WealthEngine()
allocation = engine.get_allocation("张伟")
print(allocation.to_markdown())
Enter fullscreen mode Exit fullscreen mode

The wealth-management package has the most Skills of all 7 packages (36), covering the entire wealth management chain from risk profiling and asset allocation to portfolio analysis and product recommendations.

2.5 Enterprise Risk Assessment — Multi-Dimensional Risk Profiling

from risk_compliance import RiskEngine

engine = RiskEngine()
risk = engine.get_enterprise_risk("比亚迪")
print(risk.to_markdown())
Enter fullscreen mode Exit fullscreen mode

The risk-compliance package has the broadest scenario coverage (41 Skills) — from enterprise risk assessment and AML rule detection to compliance document review, forming a complete local risk rule engine.


3. Architecture: How Does It Achieve Zero API + Millisecond Response?

3.1 Three-Layer Architecture

┌──────────────────────────────────────────┐
│            CLI / API Interface Layer      │
├──────────────────────────────────────────┤
│            Skill Business Logic Layer     │
│  Rule Engine + Decision Tree + Scoring   │
│  Model + Validation Algorithms           │
├──────────────────────────────────────────┤
│            Data Adapter Layer             │
│  Mock Data (built-in) | DB Adapter (swap)│
└──────────────────────────────────────────┘
Enter fullscreen mode Exit fullscreen mode

3.2 Performance Benchmarks

Scenario Skill Execution Comparable API Speedup
Invoice Verification 3ms 200-500ms 67-167x
Budget Control 5ms 300-800ms 60-160x
Report Quick Read 8ms 1000-3000ms 125-375x
Asset Allocation 12ms 500-1500ms 42-125x
Risk Assessment 15ms 2000-5000ms 133-333x

4. Quick Start

# Clone the repo
git clone https://github.com/yuzhaopeng-up/financial-ai-skills.git
cd financial-ai-skills

# Run directly, no dependencies to install (pure standard library)
python financial_cli.py --help
Enter fullscreen mode Exit fullscreen mode

Yep, no pip install step. Because there are zero third-party dependencies.


5. IM Integration: From Skill Output to WeCom / Feishu / DingTalk

The wecom-template-card package solves the last-mile problem of pushing Skill output to IM messages:

from wecom_template_card import MarkdownCardBuilder

builder = MarkdownCardBuilder()
card = builder.from_markdown(
    title="Budget Overspend Alert",
    markdown_table=report.to_markdown(),
    platform="wecom"  # supports wecom / feishu / dingtalk
)
Enter fullscreen mode Exit fullscreen mode

Same Skill output, zero modifications needed to push to all three IM platforms.


6. Use Cases

  • Internal financial tools: No need to send sensitive data to third-party APIs
  • High-frequency batch processing: 10K+ calls per day scenarios
  • Offline / intranet environments: Deployment environments without external API access
  • MVP rapid validation: Get the logic working first, then wire up real data
  • Training & education: Built-in mock data, students can start experimenting immediately with zero setup

Agent Skills Open Source Ecosystem

financial-ai-skills is the finance-industry vertical package in the Agent Skills open source ecosystem. The entire ecosystem follows a unified Skill specification — each repo works standalone, or can be composed together.

Repo Role GitHub
financial-ai-skills Financial AI Skills Library: 104 scenarios in pure Python https://github.com/yuzhaopeng-up/financial-ai-skills
teleagent-skills General Agent Skills: 4-Phase orchestration + rule parameterization https://github.com/yuzhaopeng-up/teleagent-skills
agent-cluster-comm Multi-Agent cluster 5-layer communication architecture https://github.com/yuzhaopeng-up/agent-cluster-comm
skill-framework Skill governance framework: L0-L4 classification + YAML templates + Python tools https://github.com/yuzhaopeng-up/skill-framework
fintech-h5-demos 12 zero-dependency financial H5 dashboard demos https://github.com/yuzhaopeng-up/fintech-h5-demos

Start your zero-API financial AI journey with financial-ai-skills — clone and run, millisecond response, MIT license.

Top comments (0)