Asking the AI about your data

Every step of the Dotwave pipeline carries a built-in AI assistant that knows your dataset. Rather than hunting through the profile yourself, you can ask it a question in plain language — "how many nulls are in this column?" — or hand it a task to carry out, and it responds using the real numbers from your data. This article covers where the chat lives, the kinds of things you can ask at each step, what information the AI can actually see, and the suggestion chips that get you started.

Where the AI chat appears

The assistant sits at the bottom of every pipeline step, below that step's tools. Wherever you are — importing, fixing nulls, preprocessing, or engineering features — the chat is right there under the controls you are already using, so you never leave the step to ask a question. Importantly, each step has its own conversation history. The discussion you have while handling missing values stays with that step, and moving to preprocessing gives you a fresh conversation scoped to the work in front of you. This keeps each thread focused and relevant to the task at hand.

What you can ask

You can ask factual questions about your data or give the AI a task to perform. The most useful questions tend to track the step you are on:

• Step 1 (Import): "How many nulls in total?"
• Step 2 (Missing values): "Which column has the most missing values?"
• Step 3 (Preprocess): "Check for outliers in price_col."
• Step 4 (Feature engineering): "Create a profit margin column."

Beyond questions, you can phrase a request as an instruction — for example, "Fill nulls in age with the median." When you ask the AI to change something, it does not act immediately; it proposes the operation for you to confirm first, which is covered in the article on AI-proposed operations.

What the AI uses

The assistant answers from your dataset's full profile, not from guesswork. That profile includes the column names, their data types, null counts per column, summary statistics, and outlier counts — the complete structural picture of your data. Alongside the profile, the AI is given the first 3 rows of your dataset so it can see the shape of real values. What it does not receive is every row of your data. It reasons over the aggregate profile plus that small sample, which is why it can tell you precise counts and statistics without needing to read your entire table.

Suggestion chips

When a chat is empty, Dotwave shows clickable suggestion chips tailored to your current step. They are a fast way to see the kinds of questions that make sense where you are — click one to send it as your message, or type your own. The suggestions change from step to step, mirroring the tasks that matter at each stage of the pipeline, so a blank chat is never a blank page.

Treat the assistant as a knowledgeable partner sitting beside your data. Ask it to confirm a count before you act, to point you at the worst column, or to carry out a routine fix — and because its answers come straight from the profile, you can trust the figures it quotes as you make cleaning decisions.