Initial commit: CI Agent competitive-intelligence monitoring app

FastAPI + Celery + Next.js + Postgres/Redis app with company monitoring,
source collection, LLM-based change analysis, enrichment, and account
security (Turnstile, escalating lockout, email verification).

apps/api/app/prompts/relevance.py

@@ -0,0 +1,51 @@
+"""Task A: document relevance. Decides whether a collected document is
+actually about the target company and matches the user's stated focus,
+before it's used as evidence for anything else."""
+
+from __future__ import annotations
+
+from pydantic import BaseModel, Field
+
+from app.analysis.llm.base import LLMProvider
+from app.prompts.base import build_user_prompt
+
+SYSTEM_PROMPT = (
+    "You are a competitive intelligence analyst. Given a single document and a company "
+    "profile, judge only whether the document is genuinely about that company and whether "
+    "it relates to the user's stated monitoring focus. Do not summarize the document's "
+    "content here - a later task does that. Be conservative: if the document could be about "
+    "a different company with a similar name, say so."
+)
+
+
+class RelevanceAssessment(BaseModel):
+    is_relevant: bool = Field(description="Is this document genuinely about the target company?")
+    matches_focus: bool = Field(description="Does it relate to the user's stated monitoring focus?")
+    topic_categories: list[str] = Field(default_factory=list)
+    source_reliability: float = Field(ge=0.0, le=1.0)
+    reasoning: str
+
+
+async def assess_relevance(
+    llm: LLMProvider,
+    *,
+    company_name: str,
+    company_aliases: list[str],
+    monitoring_focus: str | None,
+    document_title: str | None,
+    document_text: str,
+    document_url: str,
+) -> RelevanceAssessment:
+    evidence = {
+        "company_name": company_name,
+        "company_aliases": company_aliases,
+        "monitoring_focus": monitoring_focus,
+        "document_title": document_title,
+        "document_url": document_url,
+        "document_text": document_text[:4000],
+    }
+    user_prompt = build_user_prompt(
+        "Assess whether this document is about the target company and matches its monitoring focus.",
+        evidence,
+    )
+    return await llm.generate_structured(SYSTEM_PROMPT, user_prompt, RelevanceAssessment)