Reference Implementation: LLM Prompts for Task Extraction

Prompt Template

Extract actionable tasks from the following text.

Text:
{{ocr_text}}

Return a list of JSON objects:
[
  {
    "task": "string",
    "due_date": "YYYY-MM-DD or null",
    "priority": "low/medium/high",
    "owner": "string (if mentioned)"
  }
]

Reference Implementation: LLM Prompts for Classification

Prompt Template

Classify the following document into one of these categories:
[SCHOOL, ADMIN, FINANCE, MEDICAL, TECHNICAL, MISC]

Text:
{{ocr_text}}

Response: One word only.

JSON Schema for Structured Output

To improve reliability with local models (e.g. Qwen3-Coder-Next), use JSON Mode or Constrained Output by providing a formal schema.

Task Extraction Schema

{
  "type": "array",
  "items": {
    "type": "object",
    "properties": {
      "task": { "type": "string" },
      "due_date": { "type": ["string", "null"], "format": "date" },
      "priority": { "enum": ["low", "medium", "high"] },
      "owner": { "type": ["string", "null"] }
    },
    "required": ["task", "due_date", "priority", "owner"]
  }
}

Token-Efficiency Tip

When using local models, prefer a minimal schema. Removing the owner field or reasoning can reduce output tokens by 20-30% in high-volume ingestion workflows.

Contribution Metadata

• Confidence: high
• Last reviewed: 2026-03-01

Sources / References

• https://github.com/joanmarcriera/Home-office-automations