# data-agent

**Repository Path**: new_cloud1/data-agent

## Basic Information

- **Project Name**: data-agent
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2025-12-29
- **Last Updated**: 2026-01-05

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# pn Agent

 *úŽ AI „(pnûßÇê6í ùÝê¨vgL Python ãepn

## <¯ 8Ãy'

### (pný›
- **‡,{/S~**: ( LLM ùÄº‡cI‡,yÏ{
- **áoÐÖ**: Î^Ó„‡,-ÐÖÓ„áo
- **pnv**: ž°{< Excel VLOOKUP „hsTÍ\
- **ß¡**: GROUP BYZI SQL/Pandas Í\
- **pnïÆ**: ( matplotlib þh

### zýã
- Agent 9n`„ Bê¨ Python ã
- ãïÁFê¨gLúïý¯Œîc
- /B„e¤pnA

### ,§6
- gLM„0 LLM (,
- /	é!‹gpt-4o / gpt-4o-mini	
- ú®H(7,KÕn¤HœhÏ

### ‰hgL
- ã™±b‡öÍ\ûß(IqiÍ\
- êA¸pn“Í\Œpn
- gL…öŒ8U·

## =€ ë Ë

### ‰ÅV

```bash
# ( uv¨P	
uv pip install -e .

# ( pip
pip install -e .
```

### Mn¯ƒØÏ

ú `.env` ‡ö

```bash
OPENAI_API_KEY=your_openai_api_key_here

# ï	Langfuse ý*(ŽÑ§Õ	
LANGFUSE_PUBLIC_KEY=your_key
LANGFUSE_SECRET_KEY=your_secret
LANGFUSE_HOST=https://cloud.langfuse.com
```

### ÐL”(

```bash
streamlit run main.py
```

¿î `http://localhost:8501`  Ë(

## =Ö (:‹

### :‹ 1: Äº{

```
(7: ".Š reviews h„Äº	{(:o(ÍˆàHÄº"

Agent: ã ’
# ûÖÄºpn
reviews = ctx.query("SELECT * FROM reviews")

# šI{Ð:Í
prompt = "÷åÄº{:(:o/(Íˆ/àHÄºÄº{text}"

# yÏ( LLM {HM 10 aKÕ	
reviews_sample = reviews.head(10)
reviews_sample['category'] = ctx.call_llm_batch(
    reviews_sample['comment'].tolist(),
    prompt,
    model='gpt-4o-mini'
)

# ÝXÓœ
ctx.save_table(reviews_sample, 'reviews_classified_sample')
print(reviews_sample[['comment', 'category']])
```

### :‹ 2: hvVLOOKUP	

```
(7: "Š orders hŒ products h	 product_id sTwe"

Agent: ã ’
# ûÖ$*h
orders = ctx.query("SELECT * FROM orders")
products = ctx.query("SELECT * FROM products")

# gLæÞ¥{< VLOOKUP	
result = orders.merge(
    products[['product_id', 'product_name', 'category']],
    on='product_id',
    how='left'
)

# ÝXÓœ
ctx.save_table(result, 'orders_enriched')
print(f" òsT {len(result)} a¢Upn")
```

### :‹ 3: Äß¡

```
(7: "	0:ß¡ .Œ¢Up"

Agent: ã ’
# ûÖ .pn
sales = ctx.query("SELECT * FROM sales_data")

# ÄZ
summary = sales.groupby('region').agg({
    'amount': 'sum',
    'order_id': 'count'
}).reset_index()

summary.columns = ['region', 'total_amount', 'order_count']

# ÝXvU:
ctx.save_table(summary, 'sales_by_region')
print(summary)
```

### :‹ 4: pnïÆ

```
(7: "; *0: .„ñ¶þ"

Agent: ã ’
# ûÖß¡pn
summary = ctx.query("SELECT * FROM sales_by_region")

# Ø6ñ¶þ
plt = ctx.plot()
plt.figure(figsize=(10, 6))
plt.bar(summary['region'], summary['total_amount'])
plt.xlabel('0:')
plt.ylabel(' .')
plt.title('0: .ùÔ')
plt.xticks(rotation=45)
```

## <× ¶„¾¡

```
(7 B
   “
LangGraph Agent (ã B)
   “
 Python ã
   “
(7n¤ãïÁ	
   “
CodeExecutor (‰h™±gL)
   “
DataContext (Ð›pn“LLMïÆý›)
   “
U:Óœå×þhpn	
```

### 8ÃÄö

1. **DataContext**: pn
‡
   - `ctx.query(sql)`: gL SQL åâ
   - `ctx.save_table(df, name)`: ÝX DataFrame 0pn“
   - `ctx.call_llm_batch(texts, prompt)`: yÏ( LLM
   - `ctx.show_progress(current, total, msg)`: >:Û¦
   - `ctx.plot()`: ·Ö matplotlib ùa

2. **CodeExecutor**: ãgLÎ
   - ‰hÀå;bqiÍ\
   - 8U·Ë}„ïÐ:
   - “úU·å×Œþh

3. **DataProcessAgent**: LangGraph Agent
   - ã(7 B
   -  Python ã
   - ¡gLA

## = ‰h:6

### ã™±

b„Í\
- `import os`, `import sys`
- `eval()`, `exec()`
- `open()` ‡öÍ\
- `subprocess`, `shutil` Iûß(

A¸„!W
- `pandas`, `numpy`, `matplotlib`
- `datetime`, `json`, `re`, `math`

### gLP6

- …ööô5 Ÿ
- êý¿îpn“Œ LLM API
- @	“úý«U·Œ°U

## =° ,

### !‹	é

| !‹ | (:o | , |
|------|---------|------|
| gpt-4o | B¨áoÐÖ | ƒØ |
| gpt-4o-mini | €U{è | N |

###  sžõ

1. **H(7,KÕ**
   ```python
   #  H 10 aŒÁHœ
   sample = df.head(10)
   sample['result'] = ctx.call_llm_batch(...)
   ```

2. **	é!‹**
   ```python
   #  €Uû¡( mini
   ctx.call_llm_batch(texts, prompt, model='gpt-4o-mini')
   ```

3. **yÏ**
   ```python
   #  ty!'Ï÷B!p
   ctx.call_llm_batch(texts, prompt, batch_size=100)
   ```

## =Ê :‹pn

ûßËöú$*:‹h

**reviews**: Äºpn
- id, comment, rating

**sales_data**:  .pn
- order_id, product, amount, region, date

`ïå
 êñ„pnÇãú°h

## =à  Ñ

### ÐLKÕ

```bash
python test_agent.py
```

### yîÓ„

```
data-agent/
   main.py              # ;
   test_agent.py        # KÕ,
   pyproject.toml       # VMn
   .env                 # ¯ƒØÏ
   data_analysis.db     # DuckDB pn“
   schema_cache.json    # pn“ Schema X
   README.md           # ,‡c
```

## > !.

"ÎÐ¤ Issue Œ Pull Request

## =Ä ¸ï

MIT License