Spaces:
Sleeping
Sleeping
File size: 6,585 Bytes
e5e35a3 08b77a1 e5e35a3 cc8beab e5e35a3 357c48c e5e35a3 08b77a1 e5e35a3 357c48c cc8beab e5e35a3 6710fbe e5e35a3 357c48c e5e35a3 6710fbe e5e35a3 6710fbe e5e35a3 357c48c e5e35a3 6710fbe e5e35a3 357c48c e5e35a3 6710fbe e5e35a3 cc8beab e5e35a3 cc8beab e5e35a3 357c48c e5e35a3 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 | import logging
from dotenv import load_dotenv
from langchain_openai import ChatOpenAI
from src.core.AgentCommand import AgentCommand
from src.core.FinanceState import FinanceState
from src.core.errors import add_error
from src.core.settings import get_settings
from src.rag.KnowledgeBase import KnowledgeBase
from src.rag.TavilySearchRag import TavilySearchRag
from src.data.SemanticCache import SemanticCache
logger = logging.getLogger(__name__)
class TaxAgent(AgentCommand):
def __init__(self, state: FinanceState):
load_dotenv()
self.state = state
self.tavily = TavilySearchRag()
self.kb = KnowledgeBase()
self.semantic_cache = SemanticCache()
settings = get_settings()
self.client = ChatOpenAI(model=settings.models.agent_model)
def process(self):
query = self.state.get("user_query", "") or ""
trace_id = str(self.state.get("trace_id") or "")
logger.info("[trace=%s] TaxAgent.start query_len=%d", trace_id, len(query))
profile = self.state.get("user_profile") or {}
risk = str(profile.get("risk") or "").strip().lower()
experience = str(profile.get("experience") or "").strip().lower()
cache_query = (
"agent=tax\n"
f"risk={risk}; experience={experience}\n"
f"query={query.strip().lower()}"
).strip()
# Agent-level semantic cache (prevents cross-intent collisions).
try:
cached = self.semantic_cache.check_cache(cache_query, threshold=0.91)
except Exception as e:
cached = None
add_error(
self.state,
code="semantic_cache_error",
message=str(e),
agent="tax_agent",
)
if isinstance(cached, str) and cached.strip():
logger.info("[trace=%s] TaxAgent.semantic_cache_hit", trace_id)
self.state["response"] = cached.strip()
return self.state
logger.info("[trace=%s] TaxAgent.semantic_cache_miss", trace_id)
# KB-first for tax topics.
try:
sources = self.kb.retrieve(query, k=4, categories=["tax"])
except Exception as e:
logger.info("[trace=%s] KB retrieval unavailable: %s", trace_id, e)
sources = []
if sources:
logger.info("[trace=%s] TaxAgent.source=kb hits=%d", trace_id, len(sources))
snippets = "\n\n".join(
[
f"TITLE: {s.title}\nCATEGORY: {s.category}\nEXCERPT:\n{s.excerpt}"
for s in sources
]
)
citations = "\n".join([f"- {s.title} ({s.source_path})" for s in sources])
system = (
"You are a financial education assistant focusing on taxes. "
"You must avoid personalized tax advice; provide general education and encourage consulting a professional. "
"Use only the provided KB excerpts as truth. Always include: 'Educational only, not tax or financial advice.'"
)
prompt = f"""
User question:
{query}
KB excerpts:
{snippets}
Write a clear, general educational answer. End with a short 'Sources' list.
"""
try:
msg = self.client.invoke(
[
{"role": "system", "content": system},
{"role": "user", "content": prompt},
],
temperature=0.2,
)
text = msg.content.strip()
text += "\n\nSources:\n" + citations
self.state["response"] = text
self.state["retrieved_sources"] = [
{
"title": s.title,
"category": s.category,
"source_path": s.source_path,
}
for s in sources
]
try:
self.semantic_cache.save_to_cache(cache_query, self.state["response"])
except Exception:
pass
return self.state
except Exception as e:
logger.info("[trace=%s] LLM synthesis failed, returning excerpts: %s", trace_id, e)
self.state["response"] = (
"Educational only, not tax or financial advice.\n\n"
"I found these relevant KB excerpts:\n\n"
+ snippets
+ "\n\nSources:\n"
+ citations
)
self.state["retrieved_sources"] = [
{
"title": s.title,
"category": s.category,
"source_path": s.source_path,
}
for s in sources
]
try:
self.semantic_cache.save_to_cache(cache_query, self.state["response"])
except Exception:
pass
return self.state
# Fallback: web search
tavily_results = self.tavily.search(query)
logger.info(
"[trace=%s] TaxAgent.source=tavily results=%d",
trace_id,
len(tavily_results or []),
)
if (
isinstance(tavily_results, list)
and tavily_results
and tavily_results[0].get("error")
):
add_error(
self.state,
code="tavily_error",
message=str(tavily_results[0].get("error")),
agent="tax_agent",
)
self.state["response"] = (
"Educational only, not tax or financial advice.\n\n"
f"Error retrieving tax information: {tavily_results[0].get('error')}"
)
return self.state
response_text = "Tax information (web search):\n"
for result in tavily_results or []:
if result.get("title"):
response_text += f"Title: {result['title']}\n"
if result.get("url"):
response_text += f"URL: {result['url']}\n"
if result.get("content"):
response_text += f"Content: {result['content']}...\n"
response_text += "\n"
self.state["response"] = response_text.strip()
try:
self.semantic_cache.save_to_cache(cache_query, self.state["response"])
except Exception:
pass
return self.state
|