finance-analytics

finance-analytics

A Self-Governing, AI-Augmented Data Platform on Snowflake using Snowflake CoCo CLI's full extensibility stack.

This repository is the companion project to a 7-part Medium series on Snowflake CoCo CLI. It demonstrates how to build a governed, automated finance analytics platform from scratch using Skills, Subagents, Agent Teams, Hooks, and MCP integrations.

the business problem it solves

In regulated finance, undetected drift is expensive. A duplicate transaction key skews a regulatory filing. An over-privileged role exposes customer PII. An oversized warehouse quietly burns budget and the checks only run when a senior engineer remembers.

What's at stake

• Misstated regulatory filings
• Customer PII & access exposure
• Runaway Snowflake credit spend

---

The Series

Part	Title	Link
1	Build a Production-Ready dbt Project	Read →
2	Build Custom Skills	Read →
3	Build Custom Subagents	Read →
4	Build and Run Agent Teams	Read →
5	Build Guardrails with Hooks	Read →
6	Build MCP Integrations	Read →
7	Ship to Production	Read →

---

What's Inside

finance-analytics/
├── .cortex/
│   ├── agents/
│   │   ├── data-quality-inspector.md   # Profiles raw tables, flags anomalies
│   │   ├── dbt-optimizer.md            # Reviews dbt project for best practices
│   │   ├── rbac-auditor.md             # Audits Snowflake roles and access grants
│   │   ├── finops-analyst.md           # Analyses credit consumption
│   │   └── issue-router.md             # Routes findings to GitHub, Jira, Slack
│   ├── skills/
│   │   ├── source-onboard/             # Automates raw source onboarding into dbt
│   │   └── dbt-model-generator/        # Generates dbt models from source tables
│   ├── hooks/
│   │   ├── block-destructive-sql.sh    # Blocks DROP/TRUNCATE/DELETE at runtime
│   │   ├── cost-guard.sh               # Prevents queries exceeding credit threshold
│   │   ├── audit-logger.sh             # Logs every tool call with timestamp
│   │   ├── pii-scanner.sh              # Scans file writes for PII patterns
│   │   └── log-mcp-call.sh             # Audit trail for all external MCP calls
│   └── settings.json                   # Hook configuration
├── finance_warehouse/                  # dbt project (3-layer architecture)
│   ├── models/
│   │   ├── staging/                    # Views over raw sources
│   │   ├── intermediate/               # Ephemeral business logic
│   │   └── marts/                      # Final tables for consumption
│   └── dbt_project.yml
├── reports/                            # Agent-generated output
├── AGENTS.md                           # Project conventions for Cortex Code
└── README.md

---

The Extensibility Stack

Layer	What It Does	Introduced
AGENTS.md	Project context and conventions	Part 1
Skills	Reusable playbooks your team can invoke	Part 2
Subagents	Autonomous specialist agents	Part 3
Agent Teams	Coordinated multi-agent execution	Part 4
Hooks	Lifecycle interceptors and policy enforcement	Part 5
MCP	External tool integration (GitHub, Jira, Slack)	Part 6

---

Prerequisites

• Snowflake account (trial works)
• Snowflake Cortex Code CLI installed
• dbt Core with dbt-snowflake adapter
• Node.js (for MCP servers via npx)

---

Getting Started

1. Clone the repo

git clone https://github.com/saurabhkr0/finance-analytics.git
cd finance-analytics

2. Set up dbt credentials

Create finance_warehouse/profiles.yml (not committed — see .gitignore):

finance_warehouse:
  target: dev
  outputs:
    dev:
      type: snowflake
      account: your_account
      user: your_user
      authenticator: externalbrowser
      role: YOUR_ROLE
      warehouse: YOUR_WAREHOUSE
      database: FINANCE_DEMO
      schema: RAW
      threads: 4

3. Install dbt packages

cd finance_warehouse
dbt deps

4. Configure MCP servers (Part 6+)

Set environment variables in your shell profile:

export GITHUB_PERSONAL_ACCESS_TOKEN="ghp_..."
export ATLASSIAN_SITE_NAME="yourcompany"
export ATLASSIAN_USER_EMAIL="your@email.com"
export ATLASSIAN_API_TOKEN="..."
export SLACK_BOT_TOKEN="xoxb-..."
export SLACK_TEAM_ID="T..."

Then add MCP servers:

cortex mcp add github npx -y @modelcontextprotocol/server-github \
  -e GITHUB_PERSONAL_ACCESS_TOKEN=$GITHUB_PERSONAL_ACCESS_TOKEN

cortex mcp add jira npx -y @aashari/mcp-server-atlassian-jira \
  -e ATLASSIAN_SITE_NAME=$ATLASSIAN_SITE_NAME \
  -e ATLASSIAN_USER_EMAIL=$ATLASSIAN_USER_EMAIL \
  -e ATLASSIAN_API_TOKEN=$ATLASSIAN_API_TOKEN

cortex mcp add slack npx -y @modelcontextprotocol/server-slack \
  -e SLACK_BOT_TOKEN=$SLACK_BOT_TOKEN \
  -e SLACK_TEAM_ID=$SLACK_TEAM_ID

5. Launch Cortex Code

cortex

---

Running the Health Check Swarm

From a Cortex Code session, run the full four-agent audit + issue routing in one prompt:

Launch a swarm of agents:
1. data-quality-inspector agent to profile FINANCE_DEMO.RAW
2. dbt-optimizer agent to review the dbt project
3. rbac-auditor agent to audit all roles and access
4. finops-analyst agent to analyze credit consumption for the last 30 days

When all four agents complete, launch the issue-router agent to
read all reports in reports/ and route findings to GitHub, Jira, and Slack.

Reports are written to reports/. GitHub issues, Jira tickets, and a Slack summary are created automatically.

---

CI/CD

This repo includes a GitHub Actions workflow (.github/workflows/cortex-ci.yml) that validates the Cortex Code stack on every pull request:

• Agent definitions present and correctly structured
• Hooks present and free of hardcoded secrets
• settings.json is valid JSON
• dbt_project.yml and AGENTS.md exist

---

Author

Saurabh Kumar — Senior Manager, Data & AI at Accenture Australia, Snowflake Data Superhero
Medium · LinkedIn

---

Licence

MIT