Phase 1: initial codes

2026-05-26 11:04:27 -04:00
parent 03bff49523
commit f49a283059
32 changed files with 4194 additions and 0 deletions
@@ -0,0 +1,50 @@
+from __future__ import annotations
+import anthropic
+
+
+INPUT_COST_PER_1M = 3.0   # claude-sonnet-4-6 $/1M tokens
+OUTPUT_COST_PER_1M = 15.0
+
+
+class ClaudeClient:
+    MODEL = "claude-sonnet-4-6"
+
+    def __init__(self, api_key: str):
+        self._client = anthropic.Anthropic(api_key=api_key)
+
+    def ask(
+        self,
+        prompt: str,
+        system: str = "",
+        max_tokens: int = 1024,
+    ) -> tuple[str, dict]:
+        messages = [{"role": "user", "content": prompt}]
+        kwargs = {"model": self.MODEL, "max_tokens": max_tokens, "messages": messages}
+        if system:
+            kwargs["system"] = system
+
+        response = self._client.messages.create(**kwargs)
+        text = response.content[0].text if response.content else ""
+        usage = {
+            "input_tokens": response.usage.input_tokens,
+            "output_tokens": response.usage.output_tokens,
+            "cost_usd": self._estimate_cost(response.usage.input_tokens, response.usage.output_tokens),
+        }
+        return text, usage
+
+    def stream_ask(self, prompt: str, system: str = "", max_tokens: int = 1024):
+        messages = [{"role": "user", "content": prompt}]
+        kwargs = {"model": self.MODEL, "max_tokens": max_tokens, "messages": messages}
+        if system:
+            kwargs["system"] = system
+
+        with self._client.messages.stream(**kwargs) as stream:
+            for text in stream.text_stream:
+                yield text
+
+    def _estimate_cost(self, input_tokens: int, output_tokens: int) -> float:
+        return (input_tokens / 1_000_000 * INPUT_COST_PER_1M) + (output_tokens / 1_000_000 * OUTPUT_COST_PER_1M)
+
+    @staticmethod
+    def estimate_tokens(text: str) -> int:
+        return max(1, len(text) // 4)
@@ -0,0 +1,94 @@
+STOCK_SUMMARY_SYSTEM = "You are a professional equity analyst. Be concise, factual, and avoid hype."
+
+STOCK_SUMMARY_PROMPT = """
+Ticker: {symbol}
+Company: {name}
+Sector: {sector}
+Industry: {industry}
+Market Cap: {market_cap}
+P/E Ratio: {pe_ratio}
+EPS: {eps}
+52W High: {week_52_high} | 52W Low: {week_52_low}
+Dividend Yield: {dividend_yield}
+
+Recent News Headlines:
+{headlines}
+
+Write a concise 3-paragraph stock summary:
+1. Business overview and recent performance
+2. Key financial metrics analysis
+3. Near-term catalysts and risks
+"""
+
+TECHNICAL_READ_SYSTEM = "You are a technical analyst specializing in chart pattern recognition."
+
+TECHNICAL_READ_PROMPT = """
+Ticker: {symbol}
+Period Analyzed: {period}
+Current Price: {price}
+SMA20: {sma20} | SMA50: {sma50}
+RSI(14): {rsi}
+MACD: {macd} | Signal: {macd_signal}
+Bollinger Bands: Upper {bb_upper} | Lower {bb_lower}
+Recent price action: {price_action}
+
+Provide a technical analysis in 2-3 paragraphs covering:
+1. Current trend, key support/resistance levels
+2. Indicator readings and what they signal
+3. Actionable technical outlook (bullish/bearish/neutral)
+"""
+
+SENTIMENT_SCORE_SYSTEM = "You are a financial sentiment analyst. Classify sentiment precisely."
+
+SENTIMENT_SCORE_PROMPT = """
+Ticker: {symbol}
+Analyze the sentiment of these news headlines:
+
+{headlines}
+
+Respond in this exact format:
+OVERALL: [BULLISH/BEARISH/NEUTRAL]
+SCORE: [0-100 where 0=extreme bearish, 50=neutral, 100=extreme bullish]
+REASONING: [2-3 sentences explaining the dominant themes]
+HEADLINE_BREAKDOWN:
+[For each headline: + or - or ~ and one line explanation]
+"""
+
+PRICE_OUTLOOK_SYSTEM = "You are a quantitative strategist. Base your outlook on data, not speculation."
+
+PRICE_OUTLOOK_PROMPT = """
+Ticker: {symbol}
+Current Price: {price}
+Technical snapshot: RSI={rsi}, trend={trend}, momentum={momentum}
+Recent news sentiment: {sentiment}
+Sector performance: {sector_perf}
+
+Provide a short-term price outlook in this format:
+1-DAY BIAS: [BULLISH/BEARISH/NEUTRAL] — [confidence %] — [one-line rationale]
+1-WEEK BIAS: [BULLISH/BEARISH/NEUTRAL] — [confidence %] — [one-line rationale]
+1-MONTH BIAS: [BULLISH/BEARISH/NEUTRAL] — [confidence %] — [one-line rationale]
+KEY RISKS: [2 bullet points]
+"""
+
+PORTFOLIO_REVIEW_SYSTEM = "You are a portfolio risk manager. Identify risks clearly and suggest actionable improvements."
+
+PORTFOLIO_REVIEW_PROMPT = """
+Portfolio Holdings:
+{holdings}
+
+Total Value: {total_value}
+Cash Position: {cash}
+Largest Position: {top_position}
+Sector Allocation: {sector_allocation}
+
+Provide a portfolio review covering:
+1. Concentration risk (any position > 20% or sector > 40%)
+2. Correlation risk (holdings that move together)
+3. Missing diversification (sectors, asset classes)
+4. Top 3 actionable recommendations
+"""
+
+CHAT_SYSTEM = """You are StockMind AI, an expert financial assistant integrated into a stock market application.
+You have access to real-time market context provided by the user. Answer questions about stocks, crypto,
+market trends, investment strategies, and financial analysis. Be concise and professional.
+Always note that your analysis is for informational purposes only and not financial advice."""
@@ -0,0 +1,93 @@
+from __future__ import annotations
+from datetime import datetime, timedelta
+from newsapi import NewsApiClient
+from ai.claude_client import ClaudeClient
+from ai.prompts import SENTIMENT_SCORE_SYSTEM, SENTIMENT_SCORE_PROMPT
+
+
+def fetch_news(symbol: str, api_key: str, page_size: int = 10) -> list[dict]:
+    try:
+        client = NewsApiClient(api_key=api_key)
+        from_date = (datetime.now() - timedelta(days=7)).strftime("%Y-%m-%d")
+        response = client.get_everything(
+            q=symbol,
+            language="en",
+            sort_by="publishedAt",
+            page_size=page_size,
+            from_param=from_date,
+        )
+        articles = response.get("articles", [])
+        return [
+            {
+                "title": a.get("title", ""),
+                "description": a.get("description", ""),
+                "url": a.get("url", ""),
+                "source": a.get("source", {}).get("name", ""),
+                "published_at": a.get("publishedAt", ""),
+            }
+            for a in articles
+            if a.get("title")
+        ]
+    except Exception as e:
+        return []
+
+
+def analyze_sentiment(symbol: str, articles: list[dict], claude: ClaudeClient) -> tuple[str, dict]:
+    if not articles:
+        return "No news available for sentiment analysis.", {}
+
+    headlines = "\n".join(
+        f"{i + 1}. {a['title']}" for i, a in enumerate(articles[:15])
+    )
+    prompt = SENTIMENT_SCORE_PROMPT.format(symbol=symbol, headlines=headlines)
+    text, usage = claude.ask(prompt, system=SENTIMENT_SCORE_SYSTEM, max_tokens=800)
+    return text, usage
+
+
+def cache_news(symbol: str, articles: list[dict], sentiment: str = "") -> None:
+    from db.database import get_session
+    from db.models import NewsCache
+    from datetime import datetime
+
+    with get_session() as session:
+        session.query(NewsCache).filter_by(symbol=symbol).delete()
+        for a in articles:
+            session.add(NewsCache(
+                symbol=symbol,
+                title=a.get("title", ""),
+                description=a.get("description", ""),
+                url=a.get("url", ""),
+                source=a.get("source", ""),
+                published_at=a.get("published_at", ""),
+                sentiment=sentiment,
+                fetched_at=datetime.utcnow(),
+            ))
+        session.commit()
+
+
+def get_cached_news(symbol: str, max_age_hours: int = 1) -> list[dict] | None:
+    from db.database import get_session
+    from db.models import NewsCache
+    from datetime import datetime, timedelta
+
+    cutoff = datetime.utcnow() - timedelta(hours=max_age_hours)
+    with get_session() as session:
+        rows = (
+            session.query(NewsCache)
+            .filter(NewsCache.symbol == symbol, NewsCache.fetched_at > cutoff)
+            .order_by(NewsCache.published_at.desc())
+            .all()
+        )
+        if not rows:
+            return None
+        return [
+            {
+                "title": r.title,
+                "description": r.description,
+                "url": r.url,
+                "source": r.source,
+                "published_at": r.published_at,
+                "sentiment": r.sentiment,
+            }
+            for r in rows
+        ]