oluinioluwa814/WealthMangementAdvisor

Wealth Management Advisor API

FastAPI service that answers financial education questions using Gemini, a local FAISS index, and CPU sentence-transformer models. It is educational software, not a licensed financial adviser.

Current features

• —FastAPI health, readiness, and streaming chat endpoints
• —Server-sent event responses from Gemini
• —Hugging Face finance dataset ingestion on first start
• —FAISS retrieval with local JSON document storage
• —CPU-only embedding and reranking
• —Basic request validation, rate limiting, CORS, and gzip support

Runtime components

Project structure

app.py                 FastAPI application
main.py                RAG, Gemini, and market-data logic
requirements.txt       Python dependencies
finance.index          Generated FAISS index
finance_docs.json      Generated document metadata

Memory and 4 GB deployments

The service is configured for CPU execution and a small dataset slice (train[:2000]). A 4 GB machine may run it, but this is not a guarantee: PyTorch, the embedding model, reranker, Python process, and first-start dataset/index build can create a high memory peak. The first start is the riskiest because it downloads models and builds the index.

For a 4 GB host:

• —Use one Uvicorn worker only.
• —Do not run multiple replicas on the same host.
• —Keep DATASET_SPLIT at train[:2000] or reduce it.
• —Build finance.index and finance_docs.json once, then reuse them.
• —Leave at least 1 GB of free disk space for model and dataset caches.
• —Monitor memory during the first start; use a 6 to 8 GB host if it is killed by the platform.

The application does not currently use Redis, PostgreSQL, JWT authentication, or a database. Do not configure those services unless you add their implementation.

API endpoints

Example request:

{"query": "How can I plan for inflation?"}

Run locally

py -3 -m venv .venv
.\.venv\Scripts\Activate.ps1
py -3 -m pip install -r requirements.txt

Set the required secret before starting:

$env:GEMINI_API_KEY = "your-key"
py -3 -m uvicorn app:app --host 0.0.0.0 --port 7860 --workers 1

For Linux or a container, use uvicorn app:app --host 0.0.0.0 --port 7860 --workers 1.

Configuration

Do not commit .env files or API keys. Configure secrets through the deployment platform.

Deployment

For Hugging Face Spaces, use a Docker Space or another ASGI-capable deployment. A Gradio Space configuration is not appropriate for this FastAPI entry point. Configure the platform to expose port 7860, provide GEMINI_API_KEY, and persist the generated FAISS files if possible.

User -> FastAPI -> safety check -> FAISS retrieval -> Gemini -> SSE response

The responses provide financial education only and should not be treated as personalized investment advice.

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference