feat: group private apps into andreknie-privat

2026-07-25 15:47:01 +02:00
parent 4c7d48ae6d
commit 2e83750cb7
109 changed files with 0 additions and 0 deletions
@@ -0,0 +1,89 @@
+"""Structured LLM prompts for document enrichment.
+
+Provides the system prompt that instructs the LLM to return structured JSON
+with entities, categories, tags, and action items.
+"""
+
+ENRICHMENT_SYSTEM_PROMPT = """\
+You are a knowledge management assistant. Analyze the following text extracted \
+from a document and return structured metadata as JSON.
+
+Return a JSON object with exactly these fields:
+- "title": A concise, descriptive title for this note (max 80 chars)
+- "category": One of "meeting", "project", "decision", "inbox"
+- "tags": List of relevant topic tags (lowercase, no # prefix)
+- "entities": List of detected entities, each with:
+  - "type": One of "person", "project", "date", "action_item"
+  - "value": The extracted text value
+  - "confidence": Float between 0.0 and 1.0 indicating detection confidence
+  - "position": Approximate character offset where entity appears (integer or null)
+- "action_items": List of action items, each with:
+  - "description": What needs to be done
+  - "assignee": Person responsible (string or null)
+  - "deadline": Date in ISO 8601 format, e.g. "2024-03-22" (string or null)
+- "summary": A 1-2 sentence summary of the document (string or null)
+
+Rules:
+- Be precise with entity detection. Only mark entities you are confident about.
+- For people, use their full name as the value.
+- For projects, use the project name as commonly referenced.
+- For dates, use ISO 8601 format (YYYY-MM-DD) as the value.
+- Assign confidence scores honestly: 0.9+ for clearly stated entities, 0.7-0.9 for \
+likely entities, below 0.7 for uncertain ones.
+- Category "meeting" for meeting notes, agendas, minutes.
+- Category "project" for project-specific documentation.
+- Category "decision" for recorded decisions, ADRs.
+- Category "inbox" for anything that doesn't clearly fit the above.
+
+Context: This is a personal knowledge base for a project manager working across \
+multiple business projects.{context_section}
+"""
+
+STRICT_RETRY_PROMPT = """\
+Your previous response was not valid JSON. Please respond with ONLY a valid JSON \
+object matching the schema described in the system prompt. No markdown, no code fences, \
+no explanation — just the raw JSON object.
+"""
+
+
+def build_system_prompt(
+    known_projects: list[str] | None = None,
+    known_people: list[str] | None = None,
+) -> str:
+    """Build the system prompt with optional context about known entities.
+
+    Args:
+        known_projects: List of known project names for context.
+        known_people: List of known people names for context.
+
+    Returns:
+        Complete system prompt string.
+    """
+    context_parts: list[str] = []
+
+    if known_projects:
+        projects_str = ", ".join(known_projects)
+        context_parts.append(f"Known projects: {projects_str}.")
+
+    if known_people:
+        people_str = ", ".join(known_people)
+        context_parts.append(f"Known people: {people_str}.")
+
+    if context_parts:
+        context_section = "\n" + " ".join(context_parts)
+    else:
+        context_section = ""
+
+    return ENRICHMENT_SYSTEM_PROMPT.format(context_section=context_section)
+
+
+def build_user_prompt(extracted_text: str) -> str:
+    """Build the user message containing the extracted document text.
+
+    Args:
+        extracted_text: The raw text extracted from the document.
+
+    Returns:
+        User prompt string.
+    """
+    return f"Analyze this document and return the structured JSON:\n\n{extracted_text}"