Watch
1
0
Fork
You've already forked arc
0
forked from hesabix/arc
arc/hesabixAPI/app/services/ai/ai_memory_structured.py

222 lines
7.1 KiB
Python
Raw Permalink Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

"""
حافظهٔ ساخت‌یافته (JSON) — مکمل فیلد متنی آزاد.
"""
from __future__ import annotations
import json
import re
from typing import Any, Dict, List, Optional
STRUCTURED_SCHEMA_VERSION = 1
DEFAULT_STRUCTURED: Dict[str, Any] = {
"schema_version": STRUCTURED_SCHEMA_VERSION,
"sales_goal_monthly": None,
"sales_goal_unit": "toman",
"currency_display": "toman",
"report_style": "summary",
"preferred_language": "fa",
"business_role": None,
"internal_terms": [],
"knowledge_hints": [],
}
ALLOWED_CURRENCY = {"toman", "rial", "usd", "eur"}
ALLOWED_REPORT_STYLE = {"summary", "table", "detailed"}
ALLOWED_LANGUAGE = {"fa", "en"}
def _empty_structured() -> Dict[str, Any]:
return json.loads(json.dumps(DEFAULT_STRUCTURED))
def parse_structured(raw: Any) -> Dict[str, Any]:
if raw is None or raw == "":
return _empty_structured()
if isinstance(raw, dict):
data = dict(raw)
elif isinstance(raw, str):
try:
data = json.loads(raw)
except json.JSONDecodeError:
return _empty_structured()
if not isinstance(data, dict):
return _empty_structured()
else:
return _empty_structured()
out = _empty_structured()
for key in DEFAULT_STRUCTURED:
if key in data and data[key] is not None:
out[key] = data[key]
if out.get("currency_display") not in ALLOWED_CURRENCY:
out["currency_display"] = "toman"
if out.get("report_style") not in ALLOWED_REPORT_STYLE:
out["report_style"] = "summary"
if out.get("preferred_language") not in ALLOWED_LANGUAGE:
out["preferred_language"] = "fa"
terms = out.get("internal_terms")
if isinstance(terms, list):
cleaned: List[Dict[str, str]] = []
for t in terms[:30]:
if isinstance(t, dict):
term = str(t.get("term") or "").strip()[:80]
meaning = str(t.get("meaning") or "").strip()[:200]
if term:
cleaned.append({"term": term, "meaning": meaning})
elif isinstance(t, str) and t.strip():
cleaned.append({"term": t.strip()[:80], "meaning": ""})
out["internal_terms"] = cleaned
else:
out["internal_terms"] = []
hints = out.get("knowledge_hints")
if isinstance(hints, list):
out["knowledge_hints"] = [str(h).strip()[:120] for h in hints[:10] if str(h).strip()]
else:
out["knowledge_hints"] = []
goal = out.get("sales_goal_monthly")
if goal is not None:
try:
out["sales_goal_monthly"] = float(goal)
except (TypeError, ValueError):
out["sales_goal_monthly"] = None
return out
def serialize_structured(data: Dict[str, Any]) -> str:
return json.dumps(parse_structured(data), ensure_ascii=False)
def merge_structured_patch(
current: Dict[str, Any],
patch: Optional[Dict[str, Any]],
) -> Dict[str, Any]:
if not patch:
return parse_structured(current)
base = parse_structured(current)
for key, value in patch.items():
if key not in DEFAULT_STRUCTURED or value is None:
continue
if key == "internal_terms" and isinstance(value, list):
existing = {
t["term"]: t
for t in base.get("internal_terms") or []
if isinstance(t, dict) and t.get("term")
}
for t in value:
if isinstance(t, dict) and t.get("term"):
existing[str(t["term"]).strip()] = {
"term": str(t["term"]).strip()[:80],
"meaning": str(t.get("meaning") or "").strip()[:200],
}
base["internal_terms"] = list(existing.values())[:30]
else:
base[key] = value
return parse_structured(base)
def structured_to_prompt_block(structured: Dict[str, Any]) -> str:
s = parse_structured(structured)
lines: List[str] = []
goal = s.get("sales_goal_monthly")
if goal is not None and goal > 0:
unit = s.get("sales_goal_unit") or "toman"
lines.append(f"هدف فروش ماهانه: {goal:,.0f} ({unit})")
role = s.get("business_role")
if role:
lines.append(f"نقش کاربر در کسب‌وکار: {role}")
currency = s.get("currency_display")
if currency:
lines.append(f"نمایش مبالغ: {currency}")
style = s.get("report_style")
if style:
lines.append(f"سبک گزارش ترجیحی: {style}")
lang = s.get("preferred_language")
if lang == "en":
lines.append("زبان پاسخ و تحلیل ترجیحی: انگلیسی")
elif lang == "fa":
lines.append("زبان پاسخ و تحلیل ترجیحی: فارسی")
for t in s.get("internal_terms") or []:
if isinstance(t, dict) and t.get("term"):
meaning = t.get("meaning") or ""
if meaning:
lines.append(f"اصطلاح «{t['term']}»: {meaning}")
else:
lines.append(f"اصطلاح داخلی: {t['term']}")
for hint in s.get("knowledge_hints") or []:
lines.append(f"یادآوری دانشنامه: {hint}")
if not lines:
return ""
return "\n".join(lines)
def structured_to_digest_sections(structured: Dict[str, Any], content: str) -> Dict[str, Any]:
s = parse_structured(structured)
sections: List[Dict[str, str]] = []
block = structured_to_prompt_block(s)
if block:
sections.append({"title": "تنظیمات ساخت‌یافته", "body": block})
if content and content.strip():
preview = content.strip()
if len(preview) > 800:
preview = preview[:800] + "…"
sections.append({"title": "یادداشت آزاد", "body": preview})
return {
"sections": sections,
"is_empty": not block and not (content or "").strip(),
"structured": s,
}
def extract_structured_from_text(content: str) -> Dict[str, Any]:
patch: Dict[str, Any] = {}
text = (content or "").lower()
if re.search(r"تومان|toman", text):
patch["currency_display"] = "toman"
elif re.search(r"ریال|rial", text):
patch["currency_display"] = "rial"
if re.search(r"جدول|table", text):
patch["report_style"] = "table"
elif re.search(r"خلاصه|summary", text):
patch["report_style"] = "summary"
m = re.search(
r"هدف\s*(?:فروش)?\s*(?:ماه(?:انه)?)?\s*[::]?\s*([\d،,\.]+)\s*(میلیون|میلیارد|هزار)?",
content or "",
re.IGNORECASE,
)
if m:
num_raw = m.group(1).replace("،", "").replace(",", "")
try:
val = float(num_raw)
mult = m.group(2) or ""
if "میلیارد" in mult:
val *= 1_000_000_000
elif "میلیون" in mult:
val *= 1_000_000
elif "هزار" in mult:
val *= 1_000
patch["sales_goal_monthly"] = val
patch["sales_goal_unit"] = patch.get("currency_display") or "toman"
except ValueError:
pass
return patch