AI FinOps Radar

FinOps governance layer for enterprise AI spend. Token-level cost attribution, multi-provider price comparison, budget burn-down with tiered alerts, daily anomaly detection, monthly forecasting with confidence intervals, and department chargeback rollups.

Recruiter takeaway:

"This person built the FinOps tooling every CFO is asking the platform team for. Token-level attribution, anomaly detection, monthly forecasting, and chargeback — all as testable backend logic that can drop into a finance review."

Why This Exists

Most companies shipping AI in 2026 have the same blind spot: finance can see total spend but not why it's happening. The Anthropic invoice arrives at month-end and somebody on the platform team gets pulled into a meeting they can't prepare for. There's no chargeback model. No anomaly alerts. No forecast. No way to say "engineering's copilot drove 60% of spend, marketing's mass campaign on May 4 was a one-day anomaly, and we're on track to hit budget +12% if nothing changes."

This repo is that visibility layer. It treats AI spend like any other FinOps surface — pricing catalog, budget tracking, anomaly detection, forecasting, chargeback — except built for the LLM provider mix instead of cloud compute.

Where This Sits in the Portfolio

Repo	Surface	Question it answers
mcp-sentinel	Tool calls	What MCP tools are exposed and how risky?
rag-sentinel	Retrieval	What's in the vector store and how trustworthy?
agent-codex	Decisions	Under what policies are decisions allowed?
agent-eval-arena	Pre-prod	Should this model promotion ship?
agentobserve	Runtime	What did agents actually do?
shadow-ai-detector	Egress	Who's leaking what to whom?
kinetic-flightdeck	Operator	Are we OK right now?
ai-finops-radar	Finance	Are we on budget — and why not?

Five Capabilities

1. Cost Calculator + Price Comparator

Pricing catalog covers ~18 models across Anthropic, OpenAI, Google, AWS Bedrock, Cohere, Mistral, and inference hosts (Together, Groq, Fireworks). Each entry tracks input rate, output rate, optional cached-input rate, capability tags, and context window.

Cost computation handles cached-input discounts (Anthropic-style prompt caching). Provider comparator returns ranked rows with vsBaselinePct so finance can see "switching this workload from Opus to Haiku saves 95%."

2. Budget Tracker

Per-budget evaluation returns: utilization %, days elapsed/remaining, burn rate, projected month-end spend, projected overrun, status band (healthy → caution → warning → breached), alert level, and a recommended action that names the dollar overrun.

Importantly, status combines both current utilization AND projected trajectory — so a department at 60% util on day 5 with a steep slope gets flagged as caution before it actually hits 75%.

3. Anomaly Detection

Rolling-window mean + stddev with z-score thresholds. Flags daily-grain spend outliers with severity (info / warn / critical) and rationale text. Configurable window size, z-score thresholds, and minimum absolute delta to suppress small fluctuations.

The output names what happened in dollars and percent — not just "anomaly detected."

4. Monthly Forecasting

Linear regression on day-of-month vs daily spend produces month-end forecast with 95% confidence interval computed from residual standard deviation. Falls back to mean-based projection when fewer than 4 datapoints. The output includes the slope so a CFO can see "trend is +$340/day."

5. Department Chargeback

Rollup includes per-department: total spend, share of org spend, unique users/projects/providers, top provider/model/project (with dollar contribution), cost per 1k tokens. Sorted by spend so the biggest line items are first. This is the rollup that goes into a finance review packet.

API Endpoints

Cost

Method	Endpoint	Purpose
POST	/api/cost/compute	Compute cost for tokens against a model
POST	/api/cost/compare	Multi-provider cost comparison for same workload
GET	/api/cost/catalog	Full pricing catalog
GET	/api/cost/catalog/:modelId	Single pricing entry

Budgets

Method	Endpoint	Purpose
GET	/api/budgets	List demo budgets
POST	/api/budgets/evaluate	Evaluate a budget against spend + asOf date

Insights

Method	Endpoint	Purpose
POST	/api/insights/anomalies	Detect anomalies in a daily-cost series
POST	/api/insights/forecast	Forecast month-end spend with CI
POST	/api/insights/chargeback	Department chargeback rollup

Dashboard

Method	Endpoint	Purpose
GET	/health	Service status
GET	/api/dashboard/summary	Full FinOps summary across the demo dataset

Sample: Provider Comparison

POST /api/cost/compare
{
  "inputTokens": 1000000,
  "outputTokens": 500000,
  "modelIds": ["claude-opus-4.7", "claude-sonnet-4.6", "claude-haiku-4.5", "gpt-5", "gemini-2.5-flash"]
}

{
  "rows": [
    { "provider": "Google", "modelId": "gemini-2.5-flash", "displayName": "Gemini 2.5 Flash", "tier": "mainstream", "totalCostUsd": 1.55, "vsBaselinePct": 0 },
    { "provider": "Anthropic", "modelId": "claude-haiku-4.5", "displayName": "Claude Haiku 4.5", "tier": "small", "totalCostUsd": 2.80, "vsBaselinePct": 80.6 },
    { "provider": "Anthropic", "modelId": "claude-sonnet-4.6", "displayName": "Claude Sonnet 4.6", "tier": "mainstream", "totalCostUsd": 10.50, "vsBaselinePct": 577.4 },
    { "provider": "OpenAI", "modelId": "gpt-5", "displayName": "GPT-5", "tier": "frontier", "totalCostUsd": 36.00, "vsBaselinePct": 2222.6 },
    { "provider": "Anthropic", "modelId": "claude-opus-4.7", "displayName": "Claude Opus 4.7", "tier": "frontier", "totalCostUsd": 52.50, "vsBaselinePct": 3287.1 }
  ]
}

A 20× cost difference for the same workload is the kind of number that gets executives' attention.

Operator Console Preview

Getting Started

Prerequisites

• Node.js 20+
• npm

Setup

git clone https://github.com/mizcausevic-dev/ai-finops-radar.git
cd ai-finops-radar
npm install
npm run dev

Visit:

• http://localhost:3000/health
• http://localhost:3000/api/dashboard/summary
• http://localhost:3000/api/cost/catalog

Run Tests

npm test

24 unit tests across cost calculation, provider comparison, budget evaluation, anomaly detection, forecasting, and chargeback rollup.

What This Demonstrates

• FinOps thinking applied to AI spend (the topic every director-level role asks about)
• Pricing catalog research — covers cached-input discounts, embedding pricing, inference-host alternatives
• Statistical foundation — z-score anomaly detection, linear regression forecasting with proper CI
• Budget logic that combines current AND projected utilization (not just simple percentage)
• Department-level chargeback that drives down to top model + top project per dept
• Strict-mode TypeScript with full test coverage; CI matrix on Node 20 + 22

Future Enhancements

• Pull live invoices from Anthropic / OpenAI / AWS Cost Explorer
• Multi-month historical baseline for seasonality-aware anomaly detection
• ARIMA / Prophet forecast comparison alongside linear baseline
• Budget alerts via Slack / PagerDuty webhook integration
• Org-level RBAC for chargeback access (engineering sees engineering only)
• Quarterly board-ready PDF export

Tech Stack

• Node.js, TypeScript, Express, Zod
• Helmet, CORS, Morgan
• Node test runner

Portfolio Links

• LinkedIn
• Skills Page
• Medium
• GitHub

Part of mizcausevic-dev's GitHub portfolio — AI Platform Engineering doctrine.

---

Connect: LinkedIn · Kinetic Gain · Medium · Skills