AI Assistant

The bottom-right of the home page hosts an AI Assistant — a chat bubble that opens a Groq-backed chat panel. It's available to anyone who lands on the public site (auth not required).

What it can answer

The assistant is system-prompted to be useful for:

• Time-series forecasting concepts (LSTM, ARIMA, Prophet, gradient boosting, etc.).
• Feature engineering with lag and rolling-window inputs.
• Model evaluation metrics (RMSE, MAE, MAPE, backtesting, walk-forward).
• "How do I do X in RINK?" usage questions.

What it won't do

• Run queries against your dataset — it doesn't have access to your data.
• Forecast for you — use the workspace.
• Write or execute code that touches your account.

It's an information assistant, not an action assistant.

Under the hood

Browser ─POST→ /api/ai-assistant ─→ Groq Chat Completions
                                        model=llama-3.3-70b-versatile
                                        temperature=0.7
                                        max_tokens=600

• The endpoint requires a non-empty message and ≤ 4,000 characters.
• It does not require auth — it's gated only by your Groq quota.
• Errors from Groq (rate limits, invalid keys, etc.) are passed through with their original status code.

Configuration

Two env vars on the gateway control the assistant:

Variable	Default	Notes
GROQ_API_KEY	(required)	Get one at console.groq.com
GROQ_MODEL	llama-3.3-70b-versatile	Any chat-capable Groq model

Without a key, the endpoint returns 503 and the chat shows "AI assistant is not configured on this deployment."

Disabling the assistant

To remove it entirely:

1. Don't set GROQ_API_KEY — the endpoint returns 503 and the chat widget shows an error in place of replies.
2. To hide the widget UI, remove <AIAssistant /> from Home.jsx.