# FinanceGPT ML Service — Public Technical Status

## Purpose
FinanceGPT separates its Laravel control plane from a FastAPI-based machine-learning runtime used for governed learned-model inference and evaluation. The ML service complements deterministic FinanceGPT quantitative models; it does not replace the Quant Kernel or Quant Compute controls.

## Current verification status
**Deployment verification required.** The latest internal installation run successfully installed the Python application and ML dependencies, but the post-install startup smoke test did not pass because the installed NumPy wheel required an x86-64-v2 CPU baseline not exposed by the current host. FinanceGPT therefore does not represent the affected deployment as operational until numerical imports, service startup and model-inference smoke tests pass.

## Architecture
1. FinanceGPT API / quantitative workflow
2. Governed ML gateway
3. Model registry and version policy
4. FastAPI ML service
5. Model inference / evaluation
6. Evidence and model-version lineage returned to the FinanceGPT control plane

## Capability families
- anomaly detection
- generative financial scenarios
- VAE and conditional-VAE models
- time-series generation
- governed model evaluation
- versioned inference and lineage

## Readiness gates
A deployment is considered ML-ready only when all applicable checks pass:
- Python dependency import
- NumPy numerical import
- PyTorch import and device detection
- FastAPI application startup
- health/readiness response
- at least one model inference smoke test
- model/version identification
- evidence capture

GPU runtime libraries alone do not constitute evidence that a deployment has a usable physical GPU. Hardware acceleration is verified separately from software installation.

## Security and disclosure
Public documentation intentionally omits host names, filesystem paths, credentials, private network topology and environment-specific deployment secrets.