Data Preparation API

Prefix: /api/preparation

Source: backend/app/api/preparation.py

Preview CSV

POST /api/preparation/preview

{
  "file_path": "/absolute/path/to/file.csv",
  "limit": 5
}

Returns first N rows as array of objects.

Convert CSV to JSONL

POST /api/preparation/convert

{
  "file_path": "/path/to/input.csv",
  "output_path": "/path/to/output.jsonl",
  "instruction_col": "question",
  "input_col": "context",
  "output_col": "answer"
}

input_col is optional. Maps CSV columns to instruction/input/output fields.

Returns:

{
  "rows_converted": 150,
  "rows_skipped": 3,
  "output_path": "/path/to/output.jsonl"
}

Generate MCP Dataset

POST /api/preparation/generate-mcp

{
  "model_id": "model-uuid",
  "server_id": "mcp-server-uuid",
  "prompt": "You are a helpful assistant that uses tools.",
  "output_path": "/path/to/output.jsonl"
}

Connects to the specified MCP server, discovers tools, and generates training examples. Each tool produces two entries:

1. Tool call example (instruction + tool call JSON as output)
2. Capability description (what the tool does + parameter schema)

Returns:

{
  "data": [...],
  "rows": 12
}