StockSense-AI / backend /llm_vision_analysis.py
PROSTLE's picture
fix: use absolute imports for llm_proxy
b9f2b41
Raw History Blame Contribute Delete
3.37 kB
import os
import io
import json
import base64
import matplotlib.pyplot as plt
import pandas as pd
from datetime import datetime
from dotenv import load_dotenv
load_dotenv()
from llm_proxy import get_llm_client
class LLMVisionEngine:
def __init__(self):
self.proxy = get_llm_client()
def analyze_chart(self, ticker: str, df: pd.DataFrame) -> dict:
"""
Generates a chart image and asks the LLM to interpret the patterns.
"""
if self.proxy.mode == "none":
return {"error": "LLM Proxy unavailable"}
# 1. Generate the Chart
img_bytes = self._generate_chart_bytes(ticker, df)
# 2. Prepare Prompt
prompt = f"""
Analyze this stock chart for {ticker}.
Identify key technical patterns (e.g., support/resistance, breakouts, trendlines).
Predict the likely direction for the next 5-10 bars.
Respond ONLY with this JSON format:
{{
"direction": "bullish|bearish|neutral",
"score": <float -0.99 to 0.99>,
"reasoning": "<one sentence analysis of visual patterns>",
"patterns": ["pattern1", "pattern2"]
}}
"""
try:
# 3. Send to Proxy
text = self.proxy.generate_content(prompt, image_bytes=img_bytes)
text = text.strip()
if "```" in text:
parts = text.split("```")
if len(parts) >= 2:
text = parts[1]
if text.startswith("json"):
text = text[4:]
text = text.strip()
prediction = json.loads(text)
prediction["source"] = f"llm_proxy_vision/{self.proxy.mode}"
# Ensure score is rounded and clamped
raw_score = float(prediction.get("score", 0.0))
prediction["score"] = round(max(-0.99, min(0.99, raw_score)), 3)
return prediction
except Exception as e:
print(f"[Vision] Engine error: {e}")
return {"error": str(e)}
def _generate_chart_bytes(self, ticker: str, df: pd.DataFrame) -> bytes:
"""Helper to create a matplotlib chart and return raw bytes."""
plt.figure(figsize=(10, 6))
# Plot Close price
plt.plot(df.index, df['Close'], label='Close', color='#2196F3', linewidth=2)
# Add a few key indicators if present
if 'SMA_20' in df.columns:
plt.plot(df.index, df['SMA_20'], label='SMA 20', color='#FF9800', linestyle='--', alpha=0.7)
if 'SMA_50' in df.columns:
plt.plot(df.index, df['SMA_50'], label='SMA 50', color='#E91E63', linestyle='--', alpha=0.7)
plt.title(f"{ticker} - Market Analysis Graph", color='white', fontsize=14)
plt.grid(True, alpha=0.2)
plt.legend()
# Style for Dark Mode (matching UI)
plt.gca().set_facecolor('#0f172a')
plt.gcf().set_facecolor('#0f172a')
plt.tick_params(colors='white')
for spine in plt.gca().spines.values():
spine.set_color('#334155')
# Save to buffer
buf = io.BytesIO()
plt.savefig(buf, format='png', bbox_inches='tight', facecolor='#0f172a')
plt.close()
return buf.getvalue()