Instruction file imported from pablosalvador10/hackaton-loanflow-copilot (.github/instructions/loan-origination-backend.instructions.md). Copyright stays with the author.

Loan Origination Backend Instructions

Name: loan-origination-backend
Author: pablosalvador10

Overview

The Loan Origination backend is a FastAPI application that orchestrates mortgage loan applications via an Azure AI Foundry prompt agent tailored for financial document processing and loan decisioning.

Tech Stack

Python 3.11+ with uv for dependency management
FastAPI for HTTP/SSE endpoints
Azure AI Foundry Agent SDK via the shared foundrykit library
Azure Document Intelligence for document extraction
structlog for structured logging
OpenTelemetry for distributed tracing

Key Paths

Path	Purpose
`main.py`	FastAPI app entry point (port 8002)
`api/v1/loan.py`	Router — stream, upload, application, audit, docs
`core/config.py`	`LoanOriginationSettings` (Pydantic Settings)
`models/application.py`	Domain models: Application, Document, Audit
`services/storage.py`	Storage protocol + InMemory + Cosmos
`tools/`	Tool functions: validate, extract, assemble, letter, audit
`prompts/`	System prompt for the Foundry agent

Running Locally

cd py/apps/loan-origination
uv sync
uv run uvicorn main:app --reload --host 0.0.0.0 --port 8002

Running Tests

cd py/apps/loan-origination
uv run pytest -v
uv run ruff check .

Tool Functions

All tools are synchronous and return str (JSON). This is required by the Foundry FunctionTool spec.

validate_document_quality — Uses Document Intelligence prebuilt-read to check quality
extract_document_fields — Maps document type → pre-built model for extraction
assemble_application — Computes DTI ratio and builds the application summary
generate_approval_letter — Produces a Markdown pre-approval letter
log_audit_event — Appends an event to the audit trail

Stub Mode

When DOCUMENT_INTELLIGENCE_ENDPOINT is not configured, tools 1–2 return realistic mock data. This enables local development without Azure resources.

Environment Variables

Variable	Default	Notes
`FOUNDRY_PROJECT_ENDPOINT`	—	Required for agent calls
`FOUNDRY_CREDENTIAL_MODE`	`dev`	`dev` or `managed_identity`
`STORAGE_MODE`	`inmemory`	`inmemory` or `cosmos`
`DOCUMENT_INTELLIGENCE_ENDPOINT`	—	Optional; enables real doc processing
`DOCUMENT_INTELLIGENCE_KEY`	—	Optional; if using key-based auth
`MCP_SERVER_URL`	—	Optional; enables remote MCP tool server
`MCP_SERVER_LABEL`	`lending-mcp`	Label for the MCP server
`MCP_API_KEY`	—	Optional; auth header for MCP server
`MCP_ALLOWED_TOOLS`	—	Comma-separated whitelist of MCP tool names
`MCP_APPROVAL_MODE`	`never`	`never` (auto-approve) or `always`
`API_PORT`	`8002`
`CORS_ORIGINS`	`http://localhost:5174`

Conventions

Use structlog — never print()
Keep tool functions in tools/ with one function per file
Keep API contracts in Pydantic models
Instrument tool spans with OpenTelemetry
Respect the storage abstraction in services/storage.py

{ "ocm": "1", "id": "pablosalvador10-hackaton-loanflow-copilot-loan-originati-ac11ac", "kind": "skill", "name": "loan-origination-backend", "description": "The **Loan Origination** backend is a FastAPI application that orchestrates mortgage loan applications via an Azure AI Foundry prompt agent tailored for financial document processing and loan decisioning.", "publisher": "pablosalvador10", "version": "1.0.0", "capabilities": { "domains": [ "finance" ], "tags": [ "skill-md", "github-instructions" ], "languages": [ "en" ] }, "quality_prior": 0.6, "examples": [ "The **Loan Origination** backend is a FastAPI application that orchestrates mortgage loan applications via an Azure AI Foundry prompt agent tailored for financial document processing and loan decisioning." ], "primary": false, "metadata": { "source": { "provider": "github-instructions", "repository": "https://github.com/pablosalvador10/hackaton-loanflow-copilot", "path": ".github/instructions/loan-origination-backend.instructions.md", "ref": "51ab9cb5a17d91478b6874d5eb2091223a3632b8", "url": "https://github.com/pablosalvador10/hackaton-loanflow-copilot/blob/51ab9cb5a17d91478b6874d5eb2091223a3632b8/.github/instructions/loan-origination-backend.instructions.md", "key": "pablosalvador10/hackaton-loanflow-copilot/.github/instructions/loan-origination-backend.instructions.md" }, "applies_to": "py/apps/loan-origination/**" }, "instructions": "# Loan Origination Backend Instructions\n\n## Overview\n\nThe **Loan Origination** backend is a FastAPI application that orchestrates mortgage\nloan applications via an Azure AI Foundry prompt agent tailored for financial\ndocument processing and loan decisioning.\n\n## Tech Stack\n\n- **Python 3.11+** with `uv` for dependency management\n- **FastAPI** for HTTP/SSE endpoints\n- **Azure AI Foundry Agent SDK** via the shared `foundrykit` library\n- **Azure Document Intelligence** for document extraction\n- **structlog** for structured logging\n- **OpenTelemetry** for distributed tracing\n\n## Key Paths\n\n| Path |", "cost": { "context_tokens": 774 } }

loan-origination-backend

About