Upload 14 files

import our solution

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+weather.db filter=lfs diff=lfs merge=lfs -text

agent.py

@@ -0,0 +1,33 @@
+from typing import List, Dict, Optional
+from langgraph.graph import StateGraph, END
+from pydantic import BaseModel
+from langchain_openai import ChatOpenAI
+from dotenv import load_dotenv
+from os import getenv
+from utils import AgentState
+from agents_utils import make_generate_sql_node, run_sql_node, make_generate_summary_node
+
+load_dotenv()
+
+
+llm = ChatOpenAI(
+    base_url="https://openrouter.ai/api/v1",
+    api_key=getenv("OPENROUTER_API_KEY"),
+    model="openai/gpt-oss-20b:free"
+) 
+
+generate_sql_node = make_generate_sql_node(llm)
+generate_summary_node = make_generate_summary_node(llm)
+
+graph = StateGraph(AgentState)
+
+graph.add_node("generate_sql", generate_sql_node)
+graph.add_node("run_sql", run_sql_node)
+graph.add_node("summary", generate_summary_node)
+
+graph.set_entry_point("generate_sql")
+graph.add_edge("generate_sql", "run_sql")
+graph.add_edge("run_sql", "summary")
+graph.set_finish_point("summary")
+
+agent = graph.compile()

agents_utils.py

@@ -0,0 +1,69 @@
+from utils import AgentState, SQLGenerationResult
+from prompt_generation import render_sql_generation_prompts, render_summary_generation_prompts
+from typing import Dict
+import sqlite3
+from config import DB_PATH
+
+
+def make_generate_sql_node(llm):
+    def generate_sql_node(state: AgentState) -> Dict:
+        system_prompt, user_prompt = render_sql_generation_prompts(state.question)
+
+        structured_llm = llm.with_structured_output(SQLGenerationResult)
+        result = structured_llm.invoke([
+            {"role": "system", "content": system_prompt},
+            {"role": "user", "content": user_prompt},
+        ])
+
+        return {
+            "sql_query": result.sql_query,
+            "reasoning": result.reasoning,
+        }
+    return generate_sql_node
+
+def make_generate_summary_node(llm):
+    def generate_summary_node(state: AgentState) -> Dict:
+        system_prompt, user_prompt = render_summary_generation_prompts(
+            question=state.question,
+            sql_query=state.sql_query,
+            rows=state.rows
+        )
+
+        result = llm.invoke([
+            {"role": "system", "content": system_prompt},
+            {"role": "user", "content": user_prompt}
+        ])
+
+        return {
+            "answer": result.content,
+        }
+
+    return generate_summary_node
+
+def run_sql_node(state: AgentState) -> Dict:
+    if not state.sql_query:
+        return {"error": "SQL query was not generated"}
+
+    try:
+        conn = sqlite3.connect(DB_PATH)
+        conn.row_factory = sqlite3.Row
+        cursor = conn.cursor()
+
+        cursor.execute(state.sql_query)
+        rows = [dict(row) for row in cursor.fetchall()]
+
+        if len(rows) > 50:
+            rows = rows[:50]
+
+        conn.close()
+
+        return {
+            "rows": rows,
+            "error": None
+        }
+
+    except Exception as e:
+        return {
+            "error": str(e),
+            "rows": None
+        }

app.py

@@ -0,0 +1,113 @@
+import streamlit as st
+import pandas as pd
+from agent import agent, AgentState 
+
+# =========================================
+# STREAMLIT SETTINGS
+# =========================================
+st.set_page_config(page_title="Weather AI Assistant", page_icon="🌤️", layout="wide")
+
+# Custom CSS
+st.markdown("""
+<style>
+body { background-color: #f0f4ff; }
+.chat-message {
+    padding: 12px; border-radius: 12px; margin-bottom: 12px;
+    max-width: 80%; line-height: 1.5;
+}
+.user-msg {
+    background: #e3edff; color: #0f1e46;
+    align-self: flex-end; text-align: right; margin-left: auto;
+}
+.assistant-msg {
+    background: #d5e8ff; color: #0b1b33;
+    align-self: flex-start; margin-right: auto;
+}
+.chat-container {
+    display: flex; flex-direction: column; gap: 10px;
+}
+</style>
+""", unsafe_allow_html=True)
+
+# =========================================
+# SESSION STATE
+# =========================================
+if "messages" not in st.session_state:
+    st.session_state.messages = []
+if "last_details" not in st.session_state:
+    st.session_state.last_details = None
+
+# =========================================
+# MAIN TITLE
+# =========================================
+st.title("🌤️ Weather Data Chat Assistant")
+st.write("Ask questions about weather data — I will generate SQL, run it, and answer.")
+
+# =========================================
+# CHAT MESSAGES RENDER
+# =========================================
+# Спочатку малюємо історію чату
+st.markdown("<div class='chat-container'>", unsafe_allow_html=True)
+for role, msg in st.session_state.messages:
+    if role == "user":
+        st.markdown(f"<div class='chat-message user-msg'><b>You:</b> {msg}</div>", unsafe_allow_html=True)
+    else:
+        st.markdown(f"<div class='chat-message assistant-msg'><b>Assistant:</b> {msg}</div>", unsafe_allow_html=True)
+st.markdown("</div>", unsafe_allow_html=True)
+
+# =========================================
+# USER INPUT & LOGIC
+# =========================================
+user_input = st.chat_input("Type your question here...")
+
+if user_input:
+    st.session_state.messages.append(("user", user_input))
+    st.rerun() 
+
+if st.session_state.messages and st.session_state.messages[-1][0] == "user":
+    last_user_msg = st.session_state.messages[-1][1]
+    
+    with st.spinner("Thinking and querying database..."):
+        try:
+            raw_state = agent.invoke({"question": last_user_msg})
+            
+            answer = raw_state.get("answer", "No answer generated.")
+            sql_query = raw_state.get("sql_query")
+            rows = raw_state.get("rows")
+            reasoning = raw_state.get("reasoning")
+
+            st.session_state.messages.append(("assistant", answer))
+            
+            st.session_state.last_details = {
+                "sql": sql_query,
+                "rows": rows,
+                "reasoning": reasoning
+            }
+            
+            st.rerun()
+
+        except Exception as e:
+            st.session_state.messages.append(("assistant", f"❌ Error: {e}"))
+            st.rerun()
+
+# =========================================
+# DEBUG / DETAILS SECTION
+# =========================================
+if st.session_state.last_details:
+    with st.expander("🔍 See Technical Details (SQL & Data)", expanded=False):
+        details = st.session_state.last_details
+        
+        if details["reasoning"]:
+            st.write("**Reasoning:**")
+            st.info(details["reasoning"])
+            
+        if details["sql"]:
+            st.write("**Generated SQL:**")
+            st.code(details["sql"], language="sql")
+        
+        if details["rows"]:
+            st.write(f"**Data Found ({len(details['rows'])} rows):**")
+            df = pd.DataFrame(details["rows"])
+            st.dataframe(df)
+        else:
+            st.warning("No data returned from SQL.")

config.py

@@ -0,0 +1,33 @@
+import os
+
+OPENROUTER_API_KEY = os.getenv("OPENROUTER_API_KEY")
+DB_PATH = 'weather.db'
+MODEL = "openai/gpt-oss-20b:free"
+
+TABLE_NAME = "weather_daily"
+START_DATE = "1980-01-01T000000Z"
+END_DATE = "2019-12-31T230000Z"
+
+COLUMNS_TO_KEEP = [
+    "utc_timestamp",
+
+    "AT_temperature",
+    "AT_radiation_direct_horizontal",
+    "AT_radiation_diffuse_horizontal",
+
+    "BE_temperature",
+    "BE_radiation_direct_horizontal",
+    "BE_radiation_diffuse_horizontal",
+
+    "BG_temperature",
+    "BG_radiation_direct_horizontal",
+    "BG_radiation_diffuse_horizontal",
+
+    "CH_temperature",
+    "CH_radiation_direct_horizontal",
+    "CH_radiation_diffuse_horizontal",
+
+    "CZ_temperature",
+    "CZ_radiation_direct_horizontal",
+    "CZ_radiation_diffuse_horizontal",
+]

database.py

@@ -0,0 +1,16 @@
+import sqlite3
+import pandas as pd
+from config import DB_PATH, COLUMNS_TO_KEEP, TABLE_NAME
+
+def init_db_from_csv(csv_path: str = "weather_data.csv"):
+    conn = sqlite3.connect(DB_PATH)
+
+    df = pd.read_csv(csv_path, usecols=COLUMNS_TO_KEEP)
+    df.to_sql(TABLE_NAME, conn, if_exists="replace", index=False)
+
+    conn.close()
+
+# init_db_from_csv()
+
+def get_connection():
+    return sqlite3.connect(DB_PATH)

prompt_generation.py

@@ -0,0 +1,37 @@
+from jinja2 import Environment, FileSystemLoader, select_autoescape, StrictUndefined
+from config import TABLE_NAME, START_DATE, END_DATE
+from typing import List, Dict
+
+env = Environment(
+    loader=FileSystemLoader("templates"),
+    autoescape=select_autoescape(disabled_extensions=("jinja2",)),
+    undefined=StrictUndefined,
+    trim_blocks=True,
+    lstrip_blocks=True,
+)
+
+system_sql_template = env.get_template("system_prompts/system_prompt_sql_generation.jinja2")
+user_sql_template = env.get_template("user_prompts/user_prompt_sql_generation.jinja2")
+
+system_summary_template = env.get_template("system_prompts/system_prompt_summary_generation.jinja2")
+user_summary_template = env.get_template("user_prompts/user_prompt_summary_generation.jinja2")
+
+def render_sql_generation_prompts(question: str) -> tuple[str, str]:
+    system_prompt = system_sql_template.render(
+        table_name=TABLE_NAME,
+        start_date=START_DATE,
+        end_date=END_DATE,
+    )
+    user_prompt = user_sql_template.render(
+        question=question
+    )
+    return system_prompt, user_prompt
+
+def render_summary_generation_prompts(question: str, sql_query: str, rows: List[Dict]) -> tuple[str, str]:
+    system_prompt = system_summary_template.render()
+    user_prompt = user_summary_template.render(
+        question=question,
+        sql_query=sql_query,
+        rows=rows
+    )
+    return system_prompt, user_prompt

requirements.txt

@@ -0,0 +1,11 @@
+streamlit
+pandas
+openai
+langchain
+langchain-core
+langchain-openai
+langgraph
+langgraph-checkpoint
+langgraph-prebuilt
+python-dotenv
+tiktoken

templates/system_prompts/system_prompt_sql_generation.jinja2

@@ -0,0 +1,65 @@
+You are an SQL generator for an SQLite weather database.
+
+** Database Information **
+   * General Information *
+     Table name: {{ table_name }}
+
+   * Available columns *
+    utc_timestamp :
+        Column_type: TEXT ISO8601.
+        Description: Start of interval in UTC (format YYYY-MM-DDTHHMMSSZ, unique per record).
+
+    AT_temperature :
+        Column_type: REAL.
+        Description: Average temperature for Austria (AT) in °C.
+    AT_radiation_direct_horizontal :
+        Column_type: REAL.
+        Description: Direct horizontal solar radiation for Austria (AT) in W/m².
+    AT_radiation_diffuse_horizontal :
+        Column_type: REAL.
+        Description: Diffuse horizontal solar radiation for Austria (AT) in W/m².
+
+    BE_temperature :
+        Column_type: REAL.
+        Description: Average temperature for Belgium (BE) in °C.
+    BE_radiation_direct_horizontal :
+        Column_type: REAL.
+        Description: Direct horizontal solar radiation for Belgium (BE) in W/m².
+    BE_radiation_diffuse_horizontal :
+        Column_type: REAL.
+        Description: Diffuse horizontal solar radiation for Belgium (BE) in W/m².
+
+    BG_temperature :
+        Column_type: REAL.
+        Description: Average temperature for Bulgaria (BG) in °C.
+    BG_radiation_direct_horizontal :
+        Column_type: REAL.
+        Description: Direct horizontal solar radiation for Bulgaria (BG) in W/m².
+    BG_radiation_diffuse_horizontal :
+        Column_type: REAL.
+        Description: Diffuse horizontal solar radiation for Bulgaria (BG) in W/m².
+
+    CH_temperature :
+        Column_type: REAL.
+        Description: Average temperature for Switzerland (CH) in °C.
+    CH_radiation_direct_horizontal :
+        Column_type: REAL.
+        Description: Direct horizontal solar radiation for Switzerland (CH) in W/m².
+    CH_radiation_diffuse_horizontal :
+        Column_type: REAL.
+        Description: Diffuse horizontal solar radiation for Switzerland (CH) in W/m².
+
+    CZ_temperature :
+        Column_type: REAL.
+        Description: Average temperature for Czechia (CZ) in °C.
+    CZ_radiation_direct_horizontal :
+        Column_type: REAL.
+        Description: Direct horizontal solar radiation for Czechia (CZ) in W/m².
+    CZ_radiation_diffuse_horizontal :
+        Column_type: REAL.
+        Description: Diffuse horizontal solar radiation for Czechia (CZ) in W/m².
+
+** SQL Query rules generation **
+- Return only SQL.
+- Ensure date filters stay within {{ start_date }} and {{ end_date }}.
+- Do not invent column names that are not listed above.

templates/system_prompts/system_prompt_summary_generation.jinja2

@@ -0,0 +1,11 @@
+You are a data analysis assistant. Your job is to create a clear and friendly summary  
+for the user based on the provided data.
+
+Guidelines:
+- Avoid SQL terminology
+- Do not mention rows, tables, or SQL queries
+- Explain insights in simple natural language
+- Be concise (3–6 sentences)
+- If the dataset is small, interpret values directly
+- If the dataset contains statistics like averages, describe them clearly
+- If data is missing, say that no information is available

templates/user_prompts/user_prompt_sql_generation.jinja2

@@ -0,0 +1,2 @@
+User question:
+{{ question }}

templates/user_prompts/user_prompt_summary_generation.jinja2

@@ -0,0 +1,10 @@
+The user asked the following question:
+{{ question }}
+
+The generated SQL query:
+{{ sql_query }}
+
+Here is the data retrieved from the database :
+{{ rows }}
+
+Write a short summary in natural language describing the result.

test.py

@@ -0,0 +1,21 @@
+from typing import List, Dict, Optional
+from langgraph.graph import StateGraph, END
+from pydantic import BaseModel
+from langchain_openai import ChatOpenAI
+from dotenv import load_dotenv
+import os
+from os import getenv
+from prompt_generation import render_sql_generation_prompts
+from utils import SQLGenerationResult
+
+load_dotenv()
+
+
+llm = ChatOpenAI(
+    base_url="https://openrouter.ai/api/v1",
+    api_key=getenv("OPENROUTER_API_KEY"),
+    model="openai/gpt-oss-20b:free"
+) 
+
+response = llm.invoke("Generate me randon SQL query")
+print(response.content)

utils.py

@@ -0,0 +1,14 @@
+from pydantic import BaseModel, Field
+from typing import Optional, List, Dict
+
+class SQLGenerationResult(BaseModel):
+    sql_query: str = Field(..., description="SQL query to execute")
+    reasoning: Optional[str] = Field(None, description="Optional explanation of the query")
+
+class AgentState(BaseModel):
+    question: str
+    sql_query: Optional[str] = None
+    reasoning: Optional[str] = None     
+    rows: Optional[List[Dict]] = None
+    answer: Optional[str] = None
+    error: Optional[str] = None

weather.db

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:accea05ffe97779e3b73b8b7e0067c96178c28658fc007556b8e376fe819f29c
+size 45793280