Build Agentic Data Workflows Using the Speciale Planner Model
Use the Speciale model to draft structured execution plans for data.
Leverage the Speciale variant to act as a 'planner' for agentic workflows, converting natural language into structured JSON plans for data analysis.
The Scenario
You have a large CSV file and need to build a structured dashboard or automated analysis report but don't know where to start the architecture.
Before & after
You would manually clean the CSV, decide on chart types, and write individual Python scripts or Excel formulas, often taking 45–90 minutes.
Ask the model to generate a structured JSON plan first, then use that plan to guide data tools, reducing the process to 5 minutes of review and execution.
The Prompt
Act as a Lead Data Analyst Planner. I have a CSV with the following columns: [COLUMN_NAMES]. My goal is to [DESCRIBE_GOAL]. Generate a structured JSON analysis plan that defines the required data cleaning steps, three specific visualization types, and the key metrics to calculate.
Speciale is tuned for reinforcement-learning-boosted reasoning, making it excellent as a 'Planner' agent that can output structured JSON to coordinate other tools or scripts.
Source
Change Log | DeepSeek API Docs"Instead of just answering questions directly, it’s designed to plan, reason, and coordinate tools for building agentic analytics systems."