mirror of
https://github.com/x1xhlol/system-prompts-and-models-of-ai-tools.git
synced 2026-06-17 23:09:35 +00:00
d8bb836614
Add revenue discovery APIs/services, launch verification gates, CI quality checks, and frontend E2E/docs updates to prepare the branch for production go-live. Made-with: Cursor
270 lines
10 KiB
Python
270 lines
10 KiB
Python
"""Revenue discovery: vertical playbooks + optional licensed search + structured LLM enrichment."""
|
|
|
|
from __future__ import annotations
|
|
|
|
import asyncio
|
|
import json
|
|
import logging
|
|
import os
|
|
from typing import Any
|
|
|
|
import httpx
|
|
from groq import AsyncGroq
|
|
|
|
from app.schemas.revenue_discovery import (
|
|
ExplorationEnrichment,
|
|
ICPBuyingCommitteeHint,
|
|
MarketSignalItem,
|
|
ProvenanceEntry,
|
|
)
|
|
from app.services.dealix_os.vertical_playbooks import get_playbook
|
|
|
|
logger = logging.getLogger("dealix.revenue_discovery")
|
|
|
|
SECTOR_TO_PLAYBOOK: dict[str, str] = {
|
|
"تقنية المعلومات": "saas_b2b",
|
|
"العقارات": "real_estate",
|
|
"الصحة": "healthcare",
|
|
"التعليم": "professional_services",
|
|
"التجزئة": "professional_services",
|
|
"المقاولات": "professional_services",
|
|
"الاستشارات": "professional_services",
|
|
"التصنيع": "professional_services",
|
|
"اللوجستيات": "professional_services",
|
|
"المالية": "professional_services",
|
|
}
|
|
|
|
|
|
def resolve_playbook_id(sector_ar: str) -> str:
|
|
return SECTOR_TO_PLAYBOOK.get(sector_ar.strip(), "saas_b2b")
|
|
|
|
|
|
async def _tavily_snippets(query: str, max_results: int = 3) -> tuple[list[dict[str, str]], list[ProvenanceEntry]]:
|
|
key = (os.getenv("TAVILY_API_KEY") or "").strip()
|
|
if not key:
|
|
return [], [
|
|
ProvenanceEntry(
|
|
field_path="market_signals",
|
|
source="unavailable",
|
|
detail="TAVILY_API_KEY not set — licensed web search skipped",
|
|
)
|
|
]
|
|
last_err: Exception | None = None
|
|
data: dict = {}
|
|
for attempt in range(3):
|
|
try:
|
|
async with httpx.AsyncClient(timeout=22.0) as client:
|
|
r = await client.post(
|
|
"https://api.tavily.com/search",
|
|
json={
|
|
"api_key": key,
|
|
"query": query,
|
|
"search_depth": "basic",
|
|
"max_results": max_results,
|
|
"include_answer": False,
|
|
},
|
|
)
|
|
r.raise_for_status()
|
|
data = r.json()
|
|
break
|
|
except Exception as e:
|
|
last_err = e
|
|
logger.warning("tavily attempt %s failed: %s", attempt + 1, e)
|
|
if attempt < 2:
|
|
await asyncio.sleep(0.4 * (attempt + 1))
|
|
else:
|
|
return [], [
|
|
ProvenanceEntry(
|
|
field_path="market_signals",
|
|
source="licensed_web_search",
|
|
detail=f"Tavily error after retries: {type(last_err).__name__ if last_err else 'unknown'}",
|
|
)
|
|
]
|
|
|
|
results = data.get("results") or []
|
|
snippets: list[dict[str, str]] = []
|
|
for item in results[:max_results]:
|
|
title = (item.get("title") or "")[:200]
|
|
content = (item.get("content") or "")[:400]
|
|
url = item.get("url") or ""
|
|
if title or content:
|
|
snippets.append({"title": title, "url": url, "content": content})
|
|
prov = [
|
|
ProvenanceEntry(
|
|
field_path="market_signals",
|
|
source="licensed_web_search",
|
|
detail="Tavily search API",
|
|
)
|
|
]
|
|
return snippets, prov
|
|
|
|
|
|
async def build_enrichment_for_lead(
|
|
*,
|
|
sector: str,
|
|
city: str,
|
|
lead: dict[str, Any],
|
|
icp_notes_ar: str = "",
|
|
icp_notes_en: str = "",
|
|
use_licensed_search: bool = True,
|
|
groq_model: str = "llama-3.1-8b-instant",
|
|
knowledge_chunks: list[dict[str, Any]] | None = None,
|
|
) -> ExplorationEnrichment:
|
|
playbook_id = resolve_playbook_id(sector)
|
|
pb = get_playbook(playbook_id) or {}
|
|
pb_label = pb.get("label_ar") or playbook_id
|
|
|
|
prov: list[ProvenanceEntry] = [
|
|
ProvenanceEntry(
|
|
field_path="vertical_playbook_id",
|
|
source="vertical_playbook_static",
|
|
detail=f"vertical_playbooks.{playbook_id}",
|
|
),
|
|
]
|
|
if icp_notes_ar or icp_notes_en:
|
|
prov.append(
|
|
ProvenanceEntry(field_path="icp_notes", source="user_input", detail="workspace ICP context")
|
|
)
|
|
|
|
k_chunks = knowledge_chunks or []
|
|
if k_chunks:
|
|
prov.append(
|
|
ProvenanceEntry(
|
|
field_path="knowledge_rag",
|
|
source="knowledge_rag",
|
|
detail=f"SectorAsset semantic search ({len(k_chunks)} chunks)",
|
|
)
|
|
)
|
|
|
|
search_snippets: list[dict[str, str]] = []
|
|
if use_licensed_search:
|
|
q = f"{lead.get('company_name','')} {sector} {city} Saudi Arabia business news"
|
|
snippets, sprov = await _tavily_snippets(q, max_results=3)
|
|
search_snippets = snippets
|
|
prov.extend(sprov)
|
|
|
|
groq_key = os.getenv("GROQ_API_KEY", "")
|
|
if not groq_key:
|
|
refs = [f"playbook:{playbook_id}"] + [f"knowledge:{c.get('title', '')}" for c in k_chunks if c.get("title")]
|
|
return ExplorationEnrichment(
|
|
vertical_playbook_id=playbook_id,
|
|
playbook_label_ar=pb_label,
|
|
icp_summary_ar="تعذر تشغيل نموذج الإثراء (GROQ_API_KEY غير مضبوط).",
|
|
partnership_angle_ar="",
|
|
rag_playbook_refs=refs,
|
|
provenance=prov
|
|
+ [ProvenanceEntry(field_path="llm", source="unavailable", detail="GROQ_API_KEY missing")],
|
|
feature_flags_used={"knowledge_rag": bool(k_chunks), "licensed_search": use_licensed_search},
|
|
)
|
|
|
|
client = AsyncGroq(api_key=groq_key)
|
|
pb_full = dict(pb) if pb else {}
|
|
pb_full["id"] = playbook_id
|
|
pb_blob = json.dumps(pb_full, ensure_ascii=False)
|
|
if len(pb_blob) > 8000:
|
|
pb_blob = pb_blob[:8000] + "\n…[truncated playbook JSON]"
|
|
search_blob = json.dumps(search_snippets, ensure_ascii=False) if search_snippets else "[]"
|
|
rag_blob = json.dumps(
|
|
[{"title": c.get("title"), "excerpt": (c.get("content") or "")[:500]} for c in k_chunks],
|
|
ensure_ascii=False,
|
|
)
|
|
|
|
prompt = f"""أنت محلل إيرادات B2B للسوق السعودي. أنتج JSON فقط حسب المخطط الذهني التالي (بالعربية حيث ينطبق).
|
|
قطاع: {sector}
|
|
مدينة: {city}
|
|
شركة مستهدفة: {json.dumps(lead, ensure_ascii=False)}
|
|
ملاحظات ICP من المستخدم (عربي): {icp_notes_ar or "—"}
|
|
ملاحظات ICP (إنجليزي اختياري): {icp_notes_en or "—"}
|
|
مقتطفات بحث مرخّص (قد تكون فارغة): {search_blob}
|
|
مقتطفات معرفة داخلية (RAG من أصول القطاع، قد تكون فارغة): {rag_blob}
|
|
معلومات playbook قطاعي كاملة (ثابتة من النظام، قد تُقتطع): {pb_blob}
|
|
|
|
أعد JSON بالشكل:
|
|
{{
|
|
"icp_summary_ar": "ملخص قصير",
|
|
"icp_summary_en": "Short EN summary for export",
|
|
"market_signals": [{{"title":"...", "summary":"...", "implication_ar":"..."}}],
|
|
"buying_committee_hints": [{{"role_ar":"...", "role_en":"...", "rationale_ar":"..."}}],
|
|
"partnership_angle_ar": "زاوية شراكة واقعية"
|
|
}}
|
|
قواعد: لا تدّعي أخباراً غير موجودة في المقتطفات؛ إن لم تتوفر أخبار، اذكر عبارات عامة قطاعية فقط. استخدم مقتطفات المعرفة الداخلية فقط كسياق إضافي عند وجودها. لا تذكر أسعاراً."""
|
|
|
|
response = await client.chat.completions.create(
|
|
model=groq_model,
|
|
messages=[{"role": "user", "content": prompt}],
|
|
temperature=0.25,
|
|
max_tokens=1200,
|
|
response_format={"type": "json_object"},
|
|
)
|
|
raw = json.loads(response.choices[0].message.content)
|
|
|
|
signals = []
|
|
for s in raw.get("market_signals") or []:
|
|
if isinstance(s, dict) and s.get("title"):
|
|
signals.append(MarketSignalItem.model_validate(s))
|
|
|
|
hints = []
|
|
for h in raw.get("buying_committee_hints") or []:
|
|
if isinstance(h, dict) and h.get("role_ar"):
|
|
hints.append(ICPBuyingCommitteeHint.model_validate(h))
|
|
|
|
prov.append(
|
|
ProvenanceEntry(
|
|
field_path="structured_enrichment",
|
|
source="llm_groq",
|
|
detail=f"model={groq_model}",
|
|
)
|
|
)
|
|
|
|
refs = [f"playbook:{playbook_id}"] + [
|
|
f"knowledge:{c.get('title', '')}" for c in k_chunks if c.get("title")
|
|
]
|
|
|
|
return ExplorationEnrichment(
|
|
vertical_playbook_id=playbook_id,
|
|
playbook_label_ar=pb_label,
|
|
icp_summary_ar=raw.get("icp_summary_ar") or "",
|
|
icp_summary_en=raw.get("icp_summary_en") or "",
|
|
market_signals=signals,
|
|
buying_committee_hints=hints,
|
|
partnership_angle_ar=raw.get("partnership_angle_ar") or "",
|
|
rag_playbook_refs=refs,
|
|
provenance=prov,
|
|
model_id=groq_model,
|
|
feature_flags_used={"knowledge_rag": bool(k_chunks), "licensed_search": use_licensed_search},
|
|
)
|
|
|
|
|
|
def attach_generation_provenance(leads: list[dict], sector: str, city: str) -> dict[str, Any]:
|
|
"""Manifest for the bulk generate response (per-field sources)."""
|
|
playbook_id = resolve_playbook_id(sector)
|
|
return {
|
|
"sector": sector,
|
|
"city": city,
|
|
"playbook_id": playbook_id,
|
|
"lead_field_sources": {
|
|
"company_name": "llm_groq",
|
|
"pain_point": "llm_groq",
|
|
"dealix_solution": "llm_groq",
|
|
"vertical_context": "vertical_playbook_static",
|
|
},
|
|
"provenance": [
|
|
ProvenanceEntry(
|
|
field_path="leads[]",
|
|
source="llm_groq",
|
|
detail="GoogleMapsLeadScraper.generate_leads_for_sector",
|
|
).model_dump(),
|
|
ProvenanceEntry(
|
|
field_path="sector_context",
|
|
source="vertical_playbook_static",
|
|
detail=f"vertical_playbooks.{playbook_id}",
|
|
).model_dump(),
|
|
],
|
|
}
|
|
|
|
|
|
def merge_persist_metadata(enrichment: ExplorationEnrichment) -> dict[str, Any]:
|
|
"""For Lead.extra_metadata merge via API metadata field."""
|
|
d = enrichment.model_dump()
|
|
return {"revenue_discovery": d, "provenance_index": [p.model_dump() for p in enrichment.provenance]}
|