A small FastAPI starter project for simple currency forecasting endpoints that use historical data in data/.

Mobile 1	Mobile 2	Mobile 3

Mobile 4	Mobile 5

Files are in the screenshot/ folder and can be replaced or extended as needed.

1. Create a virtual env and install dependencies:

./setup.sh

2. Activate the env:

source .venv/bin/activate

3. Run the dev server:

uvicorn app.main:app --reload --host 0.0.0.0 --port 8000

Open the interactive API docs: http://127.0.0.1:8000/docs

The app provides simple retrieval-augmented forecasting endpoints using historical data found in the data/ folder (≈5 years):

• GET /rag/idr-usd?days=N&amount=A — predict IDR → USD (optionally for an amount in IDR)
• GET /rag/idr-sar?days=N&amount=A — predict IDR → SAR (optionally for an amount in IDR)
• GET /rag/idr-gold-gram?days=N&amount_gram=A — predict gold price in IDR per gram (optionally for an amount_gram in grams)
• GET /rag/idr-summary — comprehensive IDR analysis summary comparing against USD, SAR, and Gold with investment recommendations

Query parameters:

• days (int): number of days ahead to predict (default 0 = today). Maximum is 7. If >7 the API returns HTTP 400 with a short message.
• amount (float): nominal amount in IDR to convert (default 1.0). Response includes predicted_for_amount showing the forecast for the provided nominal amount.
• amount_gram (float): amount in grams for gold prediction (default 1.0).

Responses are JSON with the predicted value and uncertainty bounds, for example:

{
  "pair": "idr-usd",
  "method": "prophet",
  "date": "2025-01-03",
  "predicted": 6.01234e-05,
  "predicted_lower": 5.99e-05,
  "predicted_upper": 6.03e-05,
  "trained_until": "2024-12-31"
}

The /rag/idr-summary endpoint provides a comprehensive analysis:

{
  "summary": "IDR vs USD: TURUN (-0.50%) | IDR vs SAR: TURUN (-0.51%) | Harga Emas: NAIK (+1.27%)",
  "recommendation": {
    "main_recommendation": "INVESTASI EMAS - Emas menunjukkan tren positif...",
    "best_option": "GOLD"
  }
}

If the requested date is inside the historical range, the service will return the observed value with a note.

Fetch historical data for symbols using yfinance (Yahoo Finance API, similar to TradingView):

• GET /tradingview/get/{kode}?range=R&save=S — get historical data for symbol {kode} (e.g., USDIDR, automatically adds =X for forex)
• GET /tradingview/list-code?search=S — list available ticker codes/symbols with optional search

Query parameters:

• range (str): '1y' or '5y' (default '1y')
• save (int): 1 to save data to data/{kode}_{range}.json (filename sanitized), 0 otherwise (default 0)
• search (str): optional search string to filter by code or name (case-insensitive)

Returns JSON with historical price data (Open, High, Low, Close, Volume, etc.) for the specified period. If saved, includes "saved_to" field.

The /tradingview/list-code endpoint returns a list of available tickers including Indonesian stocks, forex pairs, commodities, and cryptocurrencies.

This repository also includes a Capacitor-based mobile app located in the capacitor/ folder. The app provides a simple web UI (built with Vite) that can be packaged for Android using Capacitor.

Quick steps to build/run the Android app:

• cd capacitor
• npm install
• npm run build
• npx cap sync android
• npx cap open android # opens Android Studio

See capacitor/README.md for more details.

• Model: uses prophet (FB Prophet) and pandas to train on the data/*.json files. Models and data are cached in-memory for faster responses.
• Data: historical daily rates are in data/USDIDR_X.json, data/SAR_X.json, and data/GC_F.json. Data is automatically updated daily at 07:00 via cronjob.
• Limits: days > 7 is intentionally rejected to keep the forecasts short and reliable.
• Cronjob: The update_data.py script runs daily to fetch fresh data for GC=F, SAR=X, and USDIDR=X.

If you plan to use this in production, consider:

• Precomputing forecasts and serving from a cache or DB (cron job or background worker)
• Adding authentication, rate-limiting, and monitoring
• Using a simpler/training-once approach if you need faster startup times

The project includes an automated data update script that fetches fresh historical data daily:

• update_data.py — fetches 5-year data for GC=F (Gold), SAR=X (Saudi Riyal), and USDIDR=X (USD/IDR) using yfinance
• Cronjob is automatically set up when running ./run.sh (runs daily at 07:00)
• Data is saved to data/GC_F.json, data/SAR_X.json, and data/USDIDR_X.json

To run manually:

python update_data.py

Run the test suite:

# ensure PYTHONPATH so `app` package can be imported
export PYTHONPATH=.
source .venv/bin/activate
pytest -q

The project includes basic tests for the new /rag endpoints.

Build and run:

docker build -t llm-exchange:latest .
docker run -p 8000:8000 llm-exchange:latest

If you have a fresh Ubuntu VPS and cloned this repo, you can run the provided init script to prepare the machine, install system and Python dependencies, run tests, and create easy run/stop scripts.

Example:

# from the repository root
sudo bash init_server.sh

After the script finishes:

• Start the server: ./run.sh (runs uvicorn in background, sets up cronjob for data updates, and writes pid to uvicorn.pid and logs to uvicorn.log)
• Stop the server: ./stop.sh (stops server and removes cronjob)
• Restart the server: ./restart.sh (stops if running, then starts)
• A sample systemd unit is available at systemd/llm-exchange.service — edit it (WorkingDirectory/ExecStart/User) before copying to /etc/systemd/system/ and enabling it with systemd.

Notes:

• The init script installs system packages with apt (python3, pip, build tools, BLAS/LAPACK libs, etc). It may take several minutes depending on the machine and network.
• Prophet may take extra time to build; check uvicorn.log or the pip output if installation fails and install additional OS libs as needed.
• Cronjob for data updates is automatically set up when running ./run.sh and removed when running ./stop.sh.

If you'd like, I can also add a script that installs and enables the systemd service automatically (requires editing the unit file to set correct WorkingDirectory and User), or create an Ansible playbook for repeatable deployments.

• Name: Dian Mukti Wibowo
• GitHub: @onyet
• Email: [email protected]

This project is licensed under the MIT License - see the LICENSE file for details.