Commit ·
85b82f6
1
Parent(s): a52ea90
יצירת גישה חדשה SQL-based: SQL query generator, executor, response synthesizer, API endpoint, ו-frontend support
Browse files- .query_history.json +24 -0
- app/api.py +53 -0
- app/sql_service.py +375 -0
- app/static/app.js +112 -6
- app/static/index.html +7 -0
.query_history.json
CHANGED
|
@@ -10,5 +10,29 @@
|
|
| 10 |
"response": {
|
| 11 |
"summary": "את הנוחות של המשתמש נוחות השירות והאינטואיטיביות של הממשק שירות לקוחות על הפנים מענה זמין יותר בשירות לקוחות שירות קל וידידותי למשתמש "
|
| 12 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 13 |
}
|
| 14 |
]
|
|
|
|
| 10 |
"response": {
|
| 11 |
"summary": "את הנוחות של המשתמש נוחות השירות והאינטואיטיביות של הממשק שירות לקוחות על הפנים מענה זמין יותר בשירות לקוחות שירות קל וידידותי למשתמש "
|
| 12 |
}
|
| 13 |
+
},
|
| 14 |
+
{
|
| 15 |
+
"query": "איך המשתמשים מרגישים לגבי השירות?",
|
| 16 |
+
"response": {
|
| 17 |
+
"summary": "את הנוחות של המשתמש נוחות השירות והאינטואיטיביות של הממשק שירות לקוחות על הפנים מענה זמין יותר בשירות לקוחות שירות קל וידידותי למשתמש "
|
| 18 |
+
}
|
| 19 |
+
},
|
| 20 |
+
{
|
| 21 |
+
"query": "בלה",
|
| 22 |
+
"response": {
|
| 23 |
+
"summary": "תוגה בלה בלה בלה זריז סבבה סבבה"
|
| 24 |
+
}
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"query": "מה שלומך אחי?",
|
| 28 |
+
"response": {
|
| 29 |
+
"summary": "אלופים אתם סיבכתם עם הורה 1 הורה 2 אבתי מאוד\nקליל, פשוט וזריז אתם אלופים אתם אלופים"
|
| 30 |
+
}
|
| 31 |
+
},
|
| 32 |
+
{
|
| 33 |
+
"query": "כמה אחוזים של משתמשים אוהב לאכול שוקולד?",
|
| 34 |
+
"response": {
|
| 35 |
+
"summary": "0 משובים מכילים את הביטוי 'של משתמשים אוהב לאכול שוקולד'."
|
| 36 |
+
}
|
| 37 |
}
|
| 38 |
]
|
app/api.py
CHANGED
|
@@ -15,6 +15,7 @@ from .config import settings
|
|
| 15 |
from .data_loader import load_feedback
|
| 16 |
from .embedding import EmbeddingModel
|
| 17 |
from .rag_service import RAGService
|
|
|
|
| 18 |
from .sentiment import analyze_sentiments
|
| 19 |
from .topics import kmeans_topics
|
| 20 |
from .vector_store import FaissVectorStore
|
|
@@ -22,6 +23,7 @@ from .vector_store import FaissVectorStore
|
|
| 22 |
|
| 23 |
app = FastAPI(title="Feedback Analysis RAG Agent", version="1.0.0", default_response_class=ORJSONResponse)
|
| 24 |
svc = RAGService()
|
|
|
|
| 25 |
embedder = svc.embedder
|
| 26 |
|
| 27 |
# Simple in-memory history persisted best-effort to `.query_history.json`
|
|
@@ -55,6 +57,14 @@ class QueryResponse(BaseModel):
|
|
| 55 |
results: Optional[List[Dict[str, Any]]] = None # Optional results for frontend
|
| 56 |
|
| 57 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 58 |
@app.post("/health")
|
| 59 |
def health() -> Dict[str, str]:
|
| 60 |
"""Healthcheck endpoint.
|
|
@@ -64,6 +74,49 @@ def health() -> Dict[str, str]:
|
|
| 64 |
return {"status": "ok"}
|
| 65 |
|
| 66 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 67 |
@app.post("/ingest")
|
| 68 |
def ingest() -> Dict[str, Any]:
|
| 69 |
"""Build the vector index from Feedback.csv"""
|
|
|
|
| 15 |
from .data_loader import load_feedback
|
| 16 |
from .embedding import EmbeddingModel
|
| 17 |
from .rag_service import RAGService
|
| 18 |
+
from .sql_service import SQLFeedbackService
|
| 19 |
from .sentiment import analyze_sentiments
|
| 20 |
from .topics import kmeans_topics
|
| 21 |
from .vector_store import FaissVectorStore
|
|
|
|
| 23 |
|
| 24 |
app = FastAPI(title="Feedback Analysis RAG Agent", version="1.0.0", default_response_class=ORJSONResponse)
|
| 25 |
svc = RAGService()
|
| 26 |
+
sql_svc = SQLFeedbackService() # SQL-based service
|
| 27 |
embedder = svc.embedder
|
| 28 |
|
| 29 |
# Simple in-memory history persisted best-effort to `.query_history.json`
|
|
|
|
| 57 |
results: Optional[List[Dict[str, Any]]] = None # Optional results for frontend
|
| 58 |
|
| 59 |
|
| 60 |
+
class SQLQueryResponse(BaseModel):
|
| 61 |
+
query: str
|
| 62 |
+
summary: str
|
| 63 |
+
sql_queries: List[str]
|
| 64 |
+
query_results: List[Dict[str, Any]] # Results of SQL queries
|
| 65 |
+
visualizations: Optional[List[Dict[str, Any]]] = None
|
| 66 |
+
|
| 67 |
+
|
| 68 |
@app.post("/health")
|
| 69 |
def health() -> Dict[str, str]:
|
| 70 |
"""Healthcheck endpoint.
|
|
|
|
| 74 |
return {"status": "ok"}
|
| 75 |
|
| 76 |
|
| 77 |
+
@app.post("/query-sql", response_model=SQLQueryResponse)
|
| 78 |
+
def query_sql(req: QueryRequest) -> SQLQueryResponse:
|
| 79 |
+
"""SQL-based question answering over feedback data.
|
| 80 |
+
|
| 81 |
+
This endpoint uses a SQL-based approach:
|
| 82 |
+
1. LLM generates 1-5 SQL queries
|
| 83 |
+
2. Executes queries on feedback data
|
| 84 |
+
3. LLM synthesizes comprehensive answer
|
| 85 |
+
4. Returns answer with query results and visualizations
|
| 86 |
+
"""
|
| 87 |
+
try:
|
| 88 |
+
result = sql_svc.analyze_query(req.query)
|
| 89 |
+
|
| 90 |
+
# Convert query results to JSON-serializable format
|
| 91 |
+
query_results = []
|
| 92 |
+
for qr in result.query_results:
|
| 93 |
+
query_results.append({
|
| 94 |
+
"query": qr.query,
|
| 95 |
+
"result": qr.result.to_dict('records') if not qr.error else [],
|
| 96 |
+
"error": qr.error,
|
| 97 |
+
"row_count": len(qr.result) if not qr.error else 0
|
| 98 |
+
})
|
| 99 |
+
|
| 100 |
+
return SQLQueryResponse(
|
| 101 |
+
query=result.user_query,
|
| 102 |
+
summary=result.summary,
|
| 103 |
+
sql_queries=result.sql_queries,
|
| 104 |
+
query_results=query_results,
|
| 105 |
+
visualizations=result.visualizations
|
| 106 |
+
)
|
| 107 |
+
except Exception as e:
|
| 108 |
+
import traceback
|
| 109 |
+
error_details = traceback.format_exc()
|
| 110 |
+
print(f"Error in /query-sql endpoint: {error_details}", flush=True)
|
| 111 |
+
return SQLQueryResponse(
|
| 112 |
+
query=req.query,
|
| 113 |
+
summary=f"שגיאה: {str(e)}. אנא בדוק את הלוגים לפרטים נוספים.",
|
| 114 |
+
sql_queries=[],
|
| 115 |
+
query_results=[],
|
| 116 |
+
visualizations=None
|
| 117 |
+
)
|
| 118 |
+
|
| 119 |
+
|
| 120 |
@app.post("/ingest")
|
| 121 |
def ingest() -> Dict[str, Any]:
|
| 122 |
"""Build the vector index from Feedback.csv"""
|
app/sql_service.py
ADDED
|
@@ -0,0 +1,375 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""
|
| 2 |
+
SQL-based Feedback Analysis Service
|
| 3 |
+
|
| 4 |
+
This module implements a SQL-based approach to analyzing feedback:
|
| 5 |
+
1. LLM analyzes user query
|
| 6 |
+
2. LLM generates 1-5 SQL queries to answer the question
|
| 7 |
+
3. Execute SQL queries on the feedback DataFrame
|
| 8 |
+
4. LLM synthesizes a comprehensive answer from query + SQL queries + results
|
| 9 |
+
5. (Optional) Generate visualizations of the results
|
| 10 |
+
"""
|
| 11 |
+
|
| 12 |
+
from __future__ import annotations
|
| 13 |
+
|
| 14 |
+
import json
|
| 15 |
+
import re
|
| 16 |
+
from dataclasses import dataclass
|
| 17 |
+
from typing import List, Dict, Any, Optional
|
| 18 |
+
import pandas as pd
|
| 19 |
+
import sqlite3
|
| 20 |
+
from io import StringIO
|
| 21 |
+
|
| 22 |
+
from .config import settings
|
| 23 |
+
from .data_loader import load_feedback
|
| 24 |
+
|
| 25 |
+
try:
|
| 26 |
+
from openai import OpenAI # type: ignore
|
| 27 |
+
except Exception:
|
| 28 |
+
OpenAI = None
|
| 29 |
+
|
| 30 |
+
try:
|
| 31 |
+
import google.generativeai as genai # type: ignore
|
| 32 |
+
except Exception:
|
| 33 |
+
genai = None
|
| 34 |
+
|
| 35 |
+
|
| 36 |
+
@dataclass
|
| 37 |
+
class SQLQueryResult:
|
| 38 |
+
"""Result of a single SQL query execution."""
|
| 39 |
+
query: str
|
| 40 |
+
result: pd.DataFrame
|
| 41 |
+
error: Optional[str] = None
|
| 42 |
+
|
| 43 |
+
|
| 44 |
+
@dataclass
|
| 45 |
+
class AnalysisResult:
|
| 46 |
+
"""Complete analysis result."""
|
| 47 |
+
user_query: str
|
| 48 |
+
sql_queries: List[str]
|
| 49 |
+
query_results: List[SQLQueryResult]
|
| 50 |
+
summary: str
|
| 51 |
+
visualizations: Optional[List[Dict[str, Any]]] = None
|
| 52 |
+
|
| 53 |
+
|
| 54 |
+
class SQLFeedbackService:
|
| 55 |
+
"""Service for SQL-based feedback analysis."""
|
| 56 |
+
|
| 57 |
+
def __init__(self):
|
| 58 |
+
self.df: Optional[pd.DataFrame] = None
|
| 59 |
+
self._load_data()
|
| 60 |
+
|
| 61 |
+
def _load_data(self):
|
| 62 |
+
"""Load feedback data into memory."""
|
| 63 |
+
try:
|
| 64 |
+
self.df = load_feedback()
|
| 65 |
+
print(f"Loaded {len(self.df)} feedback records", flush=True)
|
| 66 |
+
except Exception as e:
|
| 67 |
+
print(f"Error loading feedback data: {e}", flush=True)
|
| 68 |
+
self.df = None
|
| 69 |
+
|
| 70 |
+
def _get_schema_info(self) -> str:
|
| 71 |
+
"""Get schema information for the feedback table."""
|
| 72 |
+
if self.df is None:
|
| 73 |
+
return "No data available"
|
| 74 |
+
|
| 75 |
+
schema_info = f"""
|
| 76 |
+
טבלת Feedback מכילה את השדות הבאים:
|
| 77 |
+
- ID: מזהה ייחודי של כל משוב (מספר שלם)
|
| 78 |
+
- ServiceName: שם השירות הדיגיטלי (טקסט)
|
| 79 |
+
- Level: הציון שהמשתמש נתן לשירות (מספר שלם מ-1 עד 5, כאשר 1=גרוע, 5=מעולה)
|
| 80 |
+
- Text: הטקסט החופשי שהמשתמש הזין כחלק מהפידבק (טקסט)
|
| 81 |
+
|
| 82 |
+
סטטיסטיקות כלליות:
|
| 83 |
+
- סך הכל משובים: {len(self.df)}
|
| 84 |
+
- מספר שירותים ייחודיים: {self.df['ServiceName'].nunique()}
|
| 85 |
+
- חלוקת דירוגים: {dict(self.df['Level'].value_counts().sort_index())}
|
| 86 |
+
- דירוג ממוצע: {self.df['Level'].mean():.2f}
|
| 87 |
+
"""
|
| 88 |
+
return schema_info
|
| 89 |
+
|
| 90 |
+
def analyze_query(self, query: str) -> AnalysisResult:
|
| 91 |
+
"""
|
| 92 |
+
Main analysis pipeline:
|
| 93 |
+
1. Analyze user query
|
| 94 |
+
2. Generate SQL queries
|
| 95 |
+
3. Execute SQL queries
|
| 96 |
+
4. Synthesize answer
|
| 97 |
+
"""
|
| 98 |
+
if self.df is None:
|
| 99 |
+
raise ValueError("No feedback data available. Please ensure Feedback.csv exists.")
|
| 100 |
+
|
| 101 |
+
# Step 1: Generate SQL queries
|
| 102 |
+
sql_queries = self._generate_sql_queries(query)
|
| 103 |
+
|
| 104 |
+
# Step 2: Execute SQL queries
|
| 105 |
+
query_results = self._execute_sql_queries(sql_queries)
|
| 106 |
+
|
| 107 |
+
# Step 3: Synthesize answer
|
| 108 |
+
summary = self._synthesize_answer(query, sql_queries, query_results)
|
| 109 |
+
|
| 110 |
+
# Step 4: (Optional) Generate visualizations
|
| 111 |
+
visualizations = self._generate_visualizations(query_results)
|
| 112 |
+
|
| 113 |
+
return AnalysisResult(
|
| 114 |
+
user_query=query,
|
| 115 |
+
sql_queries=sql_queries,
|
| 116 |
+
query_results=query_results,
|
| 117 |
+
summary=summary,
|
| 118 |
+
visualizations=visualizations
|
| 119 |
+
)
|
| 120 |
+
|
| 121 |
+
def _generate_sql_queries(self, query: str) -> List[str]:
|
| 122 |
+
"""
|
| 123 |
+
Use LLM to generate 1-5 SQL queries that will help answer the user's question.
|
| 124 |
+
"""
|
| 125 |
+
schema_info = self._get_schema_info()
|
| 126 |
+
|
| 127 |
+
prompt = f"""אתה אנליסט נתונים מומחה. המשתמש שאל שאלה על משובי משתמשים.
|
| 128 |
+
|
| 129 |
+
שאלת המשתמש: {query}
|
| 130 |
+
|
| 131 |
+
מידע על הטבלה:
|
| 132 |
+
{schema_info}
|
| 133 |
+
|
| 134 |
+
המשימה שלך: צור 1 עד 5 שאילתות SQL שיעזרו לענות על השאלה. כל שאילתה צריכה להיות שימושית וממוקדת.
|
| 135 |
+
|
| 136 |
+
כללים חשובים:
|
| 137 |
+
1. השתמש בשמות השדות המדויקים: ID, ServiceName, Level, Text
|
| 138 |
+
2. Level הוא מספר שלם מ-1 עד 5 (1=גרוע, 5=מעולה)
|
| 139 |
+
3. ServiceName הוא טקסט
|
| 140 |
+
4. Text הוא הטקסט החופשי של המשוב
|
| 141 |
+
5. כל שאילתה צריכה להיות תקפה SQLite
|
| 142 |
+
6. השתמש בפונקציות SQL סטנדרטיות: COUNT, AVG, GROUP BY, WHERE, LIKE, etc.
|
| 143 |
+
7. אם השאלה מתייחסת לטקסט, השתמש ב-LIKE או INSTR לחיפוש
|
| 144 |
+
8. אם השאלה מתייחסת לדירוגים, השתמש ב-Level עם תנאים מתאימים
|
| 145 |
+
9. אם השאלה מתייחסת לשירותים, השתמש ב-ServiceName
|
| 146 |
+
|
| 147 |
+
פורמט התשובה - JSON בלבד:
|
| 148 |
+
{{
|
| 149 |
+
"queries": [
|
| 150 |
+
"SELECT ...",
|
| 151 |
+
"SELECT ...",
|
| 152 |
+
...
|
| 153 |
+
]
|
| 154 |
+
}}
|
| 155 |
+
|
| 156 |
+
תן רק את ה-JSON, ללא טקסט נוסף."""
|
| 157 |
+
|
| 158 |
+
# Try Gemini first
|
| 159 |
+
if settings.gemini_api_key and genai is not None:
|
| 160 |
+
try:
|
| 161 |
+
genai.configure(api_key=settings.gemini_api_key)
|
| 162 |
+
model = genai.GenerativeModel("gemini-1.5-flash")
|
| 163 |
+
response = model.generate_content(prompt)
|
| 164 |
+
text = getattr(response, "text", None)
|
| 165 |
+
if text:
|
| 166 |
+
return self._parse_sql_queries(text)
|
| 167 |
+
except Exception as e:
|
| 168 |
+
print(f"Gemini error in SQL generation: {e}", flush=True)
|
| 169 |
+
|
| 170 |
+
# Fallback to OpenAI
|
| 171 |
+
if settings.openai_api_key and OpenAI is not None:
|
| 172 |
+
try:
|
| 173 |
+
client = OpenAI(api_key=settings.openai_api_key)
|
| 174 |
+
response = client.chat.completions.create(
|
| 175 |
+
model="gpt-4o-mini",
|
| 176 |
+
messages=[{"role": "user", "content": prompt}],
|
| 177 |
+
temperature=0.3,
|
| 178 |
+
)
|
| 179 |
+
text = response.choices[0].message.content
|
| 180 |
+
if text:
|
| 181 |
+
return self._parse_sql_queries(text)
|
| 182 |
+
except Exception as e:
|
| 183 |
+
print(f"OpenAI error in SQL generation: {e}", flush=True)
|
| 184 |
+
|
| 185 |
+
# Fallback: return empty list
|
| 186 |
+
return []
|
| 187 |
+
|
| 188 |
+
def _parse_sql_queries(self, text: str) -> List[str]:
|
| 189 |
+
"""Parse SQL queries from LLM response."""
|
| 190 |
+
# Try to extract JSON
|
| 191 |
+
try:
|
| 192 |
+
# Remove markdown code blocks if present
|
| 193 |
+
text = re.sub(r'```json\s*', '', text)
|
| 194 |
+
text = re.sub(r'```\s*', '', text)
|
| 195 |
+
text = text.strip()
|
| 196 |
+
|
| 197 |
+
# Try to parse as JSON
|
| 198 |
+
data = json.loads(text)
|
| 199 |
+
if isinstance(data, dict) and "queries" in data:
|
| 200 |
+
queries = data["queries"]
|
| 201 |
+
if isinstance(queries, list):
|
| 202 |
+
return [q for q in queries if isinstance(q, str) and q.strip()]
|
| 203 |
+
except Exception:
|
| 204 |
+
pass
|
| 205 |
+
|
| 206 |
+
# Fallback: try to extract SQL queries directly
|
| 207 |
+
sql_pattern = r'SELECT\s+.*?(?=\n\n|\nSELECT|$)'
|
| 208 |
+
matches = re.findall(sql_pattern, text, re.IGNORECASE | re.DOTALL)
|
| 209 |
+
if matches:
|
| 210 |
+
return [m.strip() for m in matches]
|
| 211 |
+
|
| 212 |
+
return []
|
| 213 |
+
|
| 214 |
+
def _execute_sql_queries(self, sql_queries: List[str]) -> List[SQLQueryResult]:
|
| 215 |
+
"""Execute SQL queries on the feedback DataFrame."""
|
| 216 |
+
if self.df is None:
|
| 217 |
+
return []
|
| 218 |
+
|
| 219 |
+
results = []
|
| 220 |
+
|
| 221 |
+
# Create in-memory SQLite database
|
| 222 |
+
conn = sqlite3.connect(':memory:')
|
| 223 |
+
try:
|
| 224 |
+
# Write DataFrame to SQLite
|
| 225 |
+
self.df.to_sql('feedback', conn, index=False, if_exists='replace')
|
| 226 |
+
|
| 227 |
+
for query in sql_queries:
|
| 228 |
+
try:
|
| 229 |
+
# Execute query
|
| 230 |
+
result_df = pd.read_sql_query(query, conn)
|
| 231 |
+
results.append(SQLQueryResult(
|
| 232 |
+
query=query,
|
| 233 |
+
result=result_df,
|
| 234 |
+
error=None
|
| 235 |
+
))
|
| 236 |
+
except Exception as e:
|
| 237 |
+
results.append(SQLQueryResult(
|
| 238 |
+
query=query,
|
| 239 |
+
result=pd.DataFrame(),
|
| 240 |
+
error=str(e)
|
| 241 |
+
))
|
| 242 |
+
finally:
|
| 243 |
+
conn.close()
|
| 244 |
+
|
| 245 |
+
return results
|
| 246 |
+
|
| 247 |
+
def _synthesize_answer(self, query: str, sql_queries: List[str],
|
| 248 |
+
query_results: List[SQLQueryResult]) -> str:
|
| 249 |
+
"""
|
| 250 |
+
Use LLM to synthesize a comprehensive answer from:
|
| 251 |
+
- User query
|
| 252 |
+
- SQL queries that were executed
|
| 253 |
+
- Results of those queries
|
| 254 |
+
"""
|
| 255 |
+
# Format query results for the prompt
|
| 256 |
+
results_text = ""
|
| 257 |
+
for i, qr in enumerate(query_results, 1):
|
| 258 |
+
results_text += f"\nשאילתה {i}:\n{qr.query}\n\n"
|
| 259 |
+
if qr.error:
|
| 260 |
+
results_text += f"שגיאה: {qr.error}\n\n"
|
| 261 |
+
else:
|
| 262 |
+
# Format result as table
|
| 263 |
+
if len(qr.result) == 0:
|
| 264 |
+
results_text += "תוצאה: אין תוצאות\n\n"
|
| 265 |
+
else:
|
| 266 |
+
results_text += f"תוצאה ({len(qr.result)} שורות):\n"
|
| 267 |
+
results_text += qr.result.to_string(index=False)
|
| 268 |
+
results_text += "\n\n"
|
| 269 |
+
|
| 270 |
+
prompt = f"""אתה אנליסט עסקי בכיר במשרד הפנים, מומחה בייעול תהליכים דיגיטליים ושיפור חוויות המשתמשים.
|
| 271 |
+
|
| 272 |
+
המשתמש שאל שאלה על משובי משתמשים על שירותים דיגיטליים.
|
| 273 |
+
|
| 274 |
+
שאלת המשתמש: {query}
|
| 275 |
+
|
| 276 |
+
כדי לענות על השאלה, בוצעו השאילתות הבאות והתקב��ו התוצאות הבאות:
|
| 277 |
+
|
| 278 |
+
{results_text}
|
| 279 |
+
|
| 280 |
+
המשימה שלך: כתוב תשובה מסכמת, ברורה ובשפה חופשית שמבוססת על התוצאות.
|
| 281 |
+
|
| 282 |
+
דרישות:
|
| 283 |
+
1. תשובה מפורטת ומקיפה (5-7 פסקאות, 400-600 מילים)
|
| 284 |
+
2. תשובה ברורה ומסודרת - לא גיבוב של מילים
|
| 285 |
+
3. כלול מספרים מדויקים מהתוצאות
|
| 286 |
+
4. הסבר את המשמעות העסקית של הממצאים
|
| 287 |
+
5. כלול המלצות מעשיות לשיפור
|
| 288 |
+
6. כתוב בעברית מקצועית וקולחת
|
| 289 |
+
7. תן תשובה שמראה הבנה עמוקה של הנתונים
|
| 290 |
+
|
| 291 |
+
מבנה התשובה:
|
| 292 |
+
1. פתיחה - סיכום מנהלים קצר (2-3 משפטים)
|
| 293 |
+
2. ניתוח מפורט של הממצאים (3-4 פסקאות)
|
| 294 |
+
3. תובנות עסקיות והמלצות (2-3 פסקאות)
|
| 295 |
+
4. סיכום (1-2 משפטים)
|
| 296 |
+
|
| 297 |
+
אם יש שגיאות בשאילתות, ציין זאת בתשובה."""
|
| 298 |
+
|
| 299 |
+
# Try Gemini first
|
| 300 |
+
if settings.gemini_api_key and genai is not None:
|
| 301 |
+
try:
|
| 302 |
+
genai.configure(api_key=settings.gemini_api_key)
|
| 303 |
+
model = genai.GenerativeModel("gemini-1.5-flash")
|
| 304 |
+
generation_config = {
|
| 305 |
+
"temperature": 0.8,
|
| 306 |
+
"top_p": 0.95,
|
| 307 |
+
"top_k": 40,
|
| 308 |
+
"max_output_tokens": 4000,
|
| 309 |
+
}
|
| 310 |
+
response = model.generate_content(prompt, generation_config=generation_config)
|
| 311 |
+
text = getattr(response, "text", None)
|
| 312 |
+
if text and text.strip():
|
| 313 |
+
return text.strip()
|
| 314 |
+
except Exception as e:
|
| 315 |
+
print(f"Gemini error in synthesis: {e}", flush=True)
|
| 316 |
+
|
| 317 |
+
# Fallback to OpenAI
|
| 318 |
+
if settings.openai_api_key and OpenAI is not None:
|
| 319 |
+
try:
|
| 320 |
+
client = OpenAI(api_key=settings.openai_api_key)
|
| 321 |
+
response = client.chat.completions.create(
|
| 322 |
+
model="gpt-4o-mini",
|
| 323 |
+
messages=[{"role": "user", "content": prompt}],
|
| 324 |
+
temperature=0.8,
|
| 325 |
+
max_tokens=3000,
|
| 326 |
+
)
|
| 327 |
+
text = response.choices[0].message.content
|
| 328 |
+
if text:
|
| 329 |
+
return text.strip()
|
| 330 |
+
except Exception as e:
|
| 331 |
+
print(f"OpenAI error in synthesis: {e}", flush=True)
|
| 332 |
+
|
| 333 |
+
# Fallback: return basic summary
|
| 334 |
+
return f"בוצעו {len(sql_queries)} שאילתות. {len([r for r in query_results if not r.error])} הצליחו."
|
| 335 |
+
|
| 336 |
+
def _generate_visualizations(self, query_results: List[SQLQueryResult]) -> Optional[List[Dict[str, Any]]]:
|
| 337 |
+
"""
|
| 338 |
+
Generate visualization specifications for query results.
|
| 339 |
+
Returns a list of visualization configs (for frontend to render).
|
| 340 |
+
"""
|
| 341 |
+
visualizations = []
|
| 342 |
+
|
| 343 |
+
for i, qr in enumerate(query_results, 1):
|
| 344 |
+
if qr.error or len(qr.result) == 0:
|
| 345 |
+
continue
|
| 346 |
+
|
| 347 |
+
# Determine visualization type based on result structure
|
| 348 |
+
result = qr.result
|
| 349 |
+
|
| 350 |
+
# If result has 2 columns, might be a bar chart
|
| 351 |
+
if len(result.columns) == 2:
|
| 352 |
+
col1, col2 = result.columns
|
| 353 |
+
# If first column is categorical and second is numeric
|
| 354 |
+
if result[col2].dtype in ['int64', 'float64']:
|
| 355 |
+
visualizations.append({
|
| 356 |
+
"type": "bar",
|
| 357 |
+
"title": f"תוצאה של שאילתה {i}",
|
| 358 |
+
"x": col1,
|
| 359 |
+
"y": col2,
|
| 360 |
+
"data": result.to_dict('records')
|
| 361 |
+
})
|
| 362 |
+
|
| 363 |
+
# If result has 1 column with numeric values, might be a distribution
|
| 364 |
+
elif len(result.columns) == 1:
|
| 365 |
+
col = result.columns[0]
|
| 366 |
+
if result[col].dtype in ['int64', 'float64']:
|
| 367 |
+
visualizations.append({
|
| 368 |
+
"type": "histogram",
|
| 369 |
+
"title": f"תוצאה של שאילתה {i}",
|
| 370 |
+
"x": col,
|
| 371 |
+
"data": result[col].tolist()
|
| 372 |
+
})
|
| 373 |
+
|
| 374 |
+
return visualizations if visualizations else None
|
| 375 |
+
|
app/static/app.js
CHANGED
|
@@ -60,7 +60,8 @@ async function sendQuery() {
|
|
| 60 |
return;
|
| 61 |
}
|
| 62 |
|
| 63 |
-
|
|
|
|
| 64 |
|
| 65 |
// Show loading state
|
| 66 |
const sendBtn = document.getElementById('send');
|
|
@@ -69,7 +70,12 @@ async function sendQuery() {
|
|
| 69 |
sendBtn.textContent = '⏳ שולח...';
|
| 70 |
|
| 71 |
try {
|
| 72 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 73 |
method: 'POST',
|
| 74 |
headers: {'Content-Type':'application/json'},
|
| 75 |
body: JSON.stringify(body)
|
|
@@ -93,16 +99,29 @@ async function sendQuery() {
|
|
| 93 |
summaryDiv.innerHTML = '<span style="color: #d32f2f;">לא התקבלה תשובה מהשרת</span>';
|
| 94 |
}
|
| 95 |
|
| 96 |
-
// Show sources if checkbox is checked
|
| 97 |
const showSources = document.getElementById('show-sources')?.checked;
|
| 98 |
const sourcesDiv = document.getElementById('resp-sources');
|
| 99 |
-
|
| 100 |
-
|
| 101 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 102 |
} else {
|
| 103 |
if (sourcesDiv) sourcesDiv.style.display = 'none';
|
| 104 |
}
|
| 105 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 106 |
// Refresh history
|
| 107 |
await refreshHistory();
|
| 108 |
|
|
@@ -121,6 +140,93 @@ async function sendQuery() {
|
|
| 121 |
}
|
| 122 |
}
|
| 123 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 124 |
function formatResponse(text) {
|
| 125 |
// Format markdown-like text (bold, lists, etc.)
|
| 126 |
let formatted = escapeHtml(text);
|
|
|
|
| 60 |
return;
|
| 61 |
}
|
| 62 |
|
| 63 |
+
// Check which approach to use
|
| 64 |
+
const approach = document.querySelector('input[name="approach"]:checked')?.value || 'sql';
|
| 65 |
|
| 66 |
// Show loading state
|
| 67 |
const sendBtn = document.getElementById('send');
|
|
|
|
| 70 |
sendBtn.textContent = '⏳ שולח...';
|
| 71 |
|
| 72 |
try {
|
| 73 |
+
let endpoint = approach === 'sql' ? '/query-sql' : '/query';
|
| 74 |
+
const body = approach === 'sql'
|
| 75 |
+
? { query: q, top_k: 5 } // top_k not used in SQL approach but kept for compatibility
|
| 76 |
+
: { query: q, top_k: 100 };
|
| 77 |
+
|
| 78 |
+
const r = await fetch(endpoint, {
|
| 79 |
method: 'POST',
|
| 80 |
headers: {'Content-Type':'application/json'},
|
| 81 |
body: JSON.stringify(body)
|
|
|
|
| 99 |
summaryDiv.innerHTML = '<span style="color: #d32f2f;">לא התקבלה תשובה מהשרת</span>';
|
| 100 |
}
|
| 101 |
|
| 102 |
+
// Show sources/query results if checkbox is checked
|
| 103 |
const showSources = document.getElementById('show-sources')?.checked;
|
| 104 |
const sourcesDiv = document.getElementById('resp-sources');
|
| 105 |
+
|
| 106 |
+
if (showSources) {
|
| 107 |
+
if (approach === 'sql' && j.query_results && j.query_results.length > 0) {
|
| 108 |
+
sourcesDiv.style.display = 'block';
|
| 109 |
+
sourcesDiv.innerHTML = formatSQLResults(j);
|
| 110 |
+
} else if (approach === 'rag' && j.results && j.results.length > 0) {
|
| 111 |
+
sourcesDiv.style.display = 'block';
|
| 112 |
+
sourcesDiv.innerHTML = formatSources(j.results);
|
| 113 |
+
} else {
|
| 114 |
+
if (sourcesDiv) sourcesDiv.style.display = 'none';
|
| 115 |
+
}
|
| 116 |
} else {
|
| 117 |
if (sourcesDiv) sourcesDiv.style.display = 'none';
|
| 118 |
}
|
| 119 |
|
| 120 |
+
// Show visualizations if SQL approach and visualizations available
|
| 121 |
+
if (approach === 'sql' && j.visualizations && j.visualizations.length > 0) {
|
| 122 |
+
showVisualizations(j.visualizations);
|
| 123 |
+
}
|
| 124 |
+
|
| 125 |
// Refresh history
|
| 126 |
await refreshHistory();
|
| 127 |
|
|
|
|
| 140 |
}
|
| 141 |
}
|
| 142 |
|
| 143 |
+
function formatSQLResults(data) {
|
| 144 |
+
let html = '<div style="margin-top: 20px;"><h4 style="color: #1976d2; margin-bottom: 12px;">שאילתות SQL שבוצעו:</h4>';
|
| 145 |
+
|
| 146 |
+
if (data.sql_queries && data.sql_queries.length > 0) {
|
| 147 |
+
html += '<div style="margin-bottom: 20px;">';
|
| 148 |
+
data.sql_queries.forEach((query, idx) => {
|
| 149 |
+
html += `<div style="background: #f5f5f5; padding: 12px; border-radius: 8px; margin-bottom: 12px; font-family: monospace; font-size: 13px; direction: ltr; text-align: left;">`;
|
| 150 |
+
html += `<strong>שאילתה ${idx + 1}:</strong><br>${escapeHtml(query)}</div>`;
|
| 151 |
+
});
|
| 152 |
+
html += '</div>';
|
| 153 |
+
}
|
| 154 |
+
|
| 155 |
+
if (data.query_results && data.query_results.length > 0) {
|
| 156 |
+
html += '<h4 style="color: #1976d2; margin-bottom: 12px;">תוצאות:</h4>';
|
| 157 |
+
data.query_results.forEach((qr, idx) => {
|
| 158 |
+
html += `<div style="margin-bottom: 20px; padding: 16px; background: #f8f9fa; border-radius: 8px;">`;
|
| 159 |
+
html += `<strong>תוצאה ${idx + 1}:</strong> `;
|
| 160 |
+
if (qr.error) {
|
| 161 |
+
html += `<span style="color: #d32f2f;">שגיאה: ${escapeHtml(qr.error)}</span>`;
|
| 162 |
+
} else {
|
| 163 |
+
html += `<span style="color: #4caf50;">${qr.row_count} שורות</span>`;
|
| 164 |
+
if (qr.result && qr.result.length > 0) {
|
| 165 |
+
html += '<table style="width: 100%; margin-top: 12px; border-collapse: collapse;">';
|
| 166 |
+
// Header
|
| 167 |
+
html += '<thead><tr style="background: #e3f2fd;">';
|
| 168 |
+
Object.keys(qr.result[0]).forEach(col => {
|
| 169 |
+
html += `<th style="padding: 8px; text-align: right; border: 1px solid #ddd;">${escapeHtml(col)}</th>`;
|
| 170 |
+
});
|
| 171 |
+
html += '</tr></thead><tbody>';
|
| 172 |
+
// Rows (limit to 10 for display)
|
| 173 |
+
qr.result.slice(0, 10).forEach(row => {
|
| 174 |
+
html += '<tr>';
|
| 175 |
+
Object.values(row).forEach(val => {
|
| 176 |
+
html += `<td style="padding: 8px; border: 1px solid #ddd;">${escapeHtml(String(val))}</td>`;
|
| 177 |
+
});
|
| 178 |
+
html += '</tr>';
|
| 179 |
+
});
|
| 180 |
+
html += '</tbody></table>';
|
| 181 |
+
if (qr.result.length > 10) {
|
| 182 |
+
html += `<div style="margin-top: 8px; color: #666; font-size: 14px;">...ועוד ${qr.result.length - 10} שורות</div>`;
|
| 183 |
+
}
|
| 184 |
+
}
|
| 185 |
+
}
|
| 186 |
+
html += '</div>';
|
| 187 |
+
});
|
| 188 |
+
}
|
| 189 |
+
|
| 190 |
+
html += '</div>';
|
| 191 |
+
return html;
|
| 192 |
+
}
|
| 193 |
+
|
| 194 |
+
function showVisualizations(visualizations) {
|
| 195 |
+
// Create or get visualizations container
|
| 196 |
+
let vizContainer = document.getElementById('resp-visualizations');
|
| 197 |
+
if (!vizContainer) {
|
| 198 |
+
vizContainer = document.createElement('div');
|
| 199 |
+
vizContainer.id = 'resp-visualizations';
|
| 200 |
+
vizContainer.style.marginTop = '20px';
|
| 201 |
+
document.getElementById('last-response').appendChild(vizContainer);
|
| 202 |
+
}
|
| 203 |
+
|
| 204 |
+
vizContainer.style.display = 'block';
|
| 205 |
+
vizContainer.innerHTML = '<h4 style="color: #1976d2; margin-bottom: 12px;">📊 גרפיקות:</h4>';
|
| 206 |
+
|
| 207 |
+
visualizations.forEach((viz, idx) => {
|
| 208 |
+
const vizDiv = document.createElement('div');
|
| 209 |
+
vizDiv.style.marginBottom = '24px';
|
| 210 |
+
vizDiv.style.padding = '16px';
|
| 211 |
+
vizDiv.style.background = '#f8f9fa';
|
| 212 |
+
vizDiv.style.borderRadius = '8px';
|
| 213 |
+
|
| 214 |
+
vizDiv.innerHTML = `<h5 style="margin-top: 0; color: #1976d2;">${viz.title}</h5>`;
|
| 215 |
+
vizDiv.innerHTML += `<div id="viz-${idx}" style="height: 300px;"></div>`;
|
| 216 |
+
|
| 217 |
+
vizContainer.appendChild(vizDiv);
|
| 218 |
+
|
| 219 |
+
// Use Chart.js or similar for visualization
|
| 220 |
+
// For now, just show a placeholder
|
| 221 |
+
document.getElementById(`viz-${idx}`).innerHTML = `
|
| 222 |
+
<div style="text-align: center; padding: 40px; color: #666;">
|
| 223 |
+
📊 גרפיקה מסוג ${viz.type}<br>
|
| 224 |
+
<small>תמיכה בגרפיקות תתווסף בקרוב</small>
|
| 225 |
+
</div>
|
| 226 |
+
`;
|
| 227 |
+
});
|
| 228 |
+
}
|
| 229 |
+
|
| 230 |
function formatResponse(text) {
|
| 231 |
// Format markdown-like text (bold, lists, etc.)
|
| 232 |
let formatted = escapeHtml(text);
|
app/static/index.html
CHANGED
|
@@ -203,6 +203,13 @@
|
|
| 203 |
<label><input type="checkbox" id="show-sources" /> הצג דוגמאות מהנתונים</label>
|
| 204 |
<span class="small" style="margin-left:12px;">ברירת מחדל: מוסתר — יוצג רק הסיכום האנליטי</span>
|
| 205 |
</div>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 206 |
<div style="display:flex;gap:8px;margin-top:12px;">
|
| 207 |
<button id="send" class="primary">🔍 שאל</button>
|
| 208 |
<button id="clear-history" class="muted">🗑️ נקה היסטוריה</button>
|
|
|
|
| 203 |
<label><input type="checkbox" id="show-sources" /> הצג דוגמאות מהנתונים</label>
|
| 204 |
<span class="small" style="margin-left:12px;">ברירת מחדל: מוסתר — יוצג רק הסיכום האנליטי</span>
|
| 205 |
</div>
|
| 206 |
+
<div style="margin-top:12px;">
|
| 207 |
+
<label style="font-weight: 600;">גישת ניתוח:</label>
|
| 208 |
+
<div style="margin-top:8px;">
|
| 209 |
+
<label><input type="radio" name="approach" value="sql" checked /> SQL-based (מומלץ - חדש)</label>
|
| 210 |
+
<label style="margin-right:20px;"><input type="radio" name="approach" value="rag" /> RAG-based (ישן)</label>
|
| 211 |
+
</div>
|
| 212 |
+
</div>
|
| 213 |
<div style="display:flex;gap:8px;margin-top:12px;">
|
| 214 |
<button id="send" class="primary">🔍 שאל</button>
|
| 215 |
<button id="clear-history" class="muted">🗑️ נקה היסטוריה</button>
|