Serverless Expense Processor Agent

A markdown-first Azure Functions serverless agent for queue-driven expense processing. Its trigger and instructions live in src/agents/expense_processor.agent.md, and Azure Functions handles execution and scale-to-zero.

What it does

• 🧾 Reads any format: text, email, key/value, or JSON, and extracts amount, currency, category, and vendor.
• 📚 Picks the right policy: lists the documents in Blob Storage and selects the one whose scope matches, then reads and applies it.
• 🚦 Routes the decision: approve → expense-approved, review → expense-review, flag / FX → expense-flagged.
• 🔀 Proves it's reasoning: the same $450 is auto-approved as travel but sent to review as a client dinner; tighten one policy document and only that category reroutes.

Prerequisites

• An Azure subscription
• uv
• Azure Developer CLI (azd)

Quickstart

azd up

Wait up to a minute, then read the decisions:

uv run scripts/read_decision.py --queue all --peek --cloud

You should see:

Request	Policy	Queue
$450 flight	travel-policy.md	expense-approved
$450 monitor	equipment-software-policy.md	expense-approved
$450 client dinner	meals-entertainment-policy.md	expense-review

Clean up with azd down --purge.

Run it locally (Azurite)

Install Azurite and Azure Functions Core Tools. Copy src/local.settings.json.sample to src/local.settings.json and set the model endpoint and deployment.

azurite --silent --location .azurite               # terminal A
cd src && uv run func start                         # terminal B
uv run scripts/send_expense.py --file samples/travel.txt   # terminal C
uv run scripts/read_decision.py --queue all --peek

The model call still uses Azure. For setup and Windows help, see Troubleshooting.

How it works

flowchart LR
    msg([raw message<br/>text · JSON · key-value])

    subgraph inbound["Azure Queue Storage · inbound"]
        inq[[expense-requests]]
    end

    subgraph policies["Azure Blob Storage · policies container"]
        pdocs[(travel · meals · equipment<br/>general policy docs)]
    end

    agent{{Expense Processor agent<br/>extract · select · apply · route}}

    subgraph outbound["Azure Queue Storage · outbound"]
        approved[[expense-approved]]
        review[[expense-review]]
        flagged[[expense-flagged]]
    end

    msg --> inq
    inq -->|queue trigger| agent
    pdocs --> agent
    agent --> approved
    agent --> review
    agent --> flagged

Loading

The runtime discovers the agent Markdown file. Its front matter defines the queue trigger, and its body contains the instructions. Three Python tools read policy documents and route decisions using managed identity.

How it works · Use cases · Customize · Deploy · Troubleshooting

Learn more

• Serverless agents runtime in Azure Functions
• Azure Functions Flex Consumption
• uv · PEP 723: inline script metadata

License

MIT © Microsoft Corporation.