101 lines
3.5 KiB
Python
101 lines
3.5 KiB
Python
import time
|
|
import json
|
|
|
|
|
|
def create_research_manager(llm, memory):
|
|
def research_manager_node(state) -> dict:
|
|
history = state["investment_debate_state"].get("history", "")
|
|
market_research_report = state["market_report"]
|
|
sentiment_report = state["sentiment_report"]
|
|
news_report = state["news_report"]
|
|
fundamentals_report = state["fundamentals_report"]
|
|
|
|
investment_debate_state = state["investment_debate_state"]
|
|
|
|
curr_situation = f"{market_research_report}\n\n{sentiment_report}\n\n{news_report}\n\n{fundamentals_report}"
|
|
|
|
if memory:
|
|
past_memories = memory.get_memories(curr_situation, n_matches=2)
|
|
else:
|
|
past_memories = []
|
|
|
|
|
|
if past_memories:
|
|
past_memory_str = "### Past Lessons Applied\\n**Reflections from Similar Situations:**\\n"
|
|
for i, rec in enumerate(past_memories, 1):
|
|
past_memory_str += rec["recommendation"] + "\\n\\n"
|
|
past_memory_str += "\\n\\n**How I'm Using These Lessons:**\\n"
|
|
past_memory_str += "- [Specific adjustment based on past mistake/success]\\n"
|
|
past_memory_str += "- [Impact on current conviction level]\\n"
|
|
else:
|
|
past_memory_str = "" # Don't include placeholder when no memories
|
|
|
|
prompt = f"""You are the Trade Judge for {state["company_of_interest"]}. Decide if there is a SHORT-TERM edge to trade this stock (1-2 weeks).
|
|
|
|
## CORE RULES (CRITICAL)
|
|
- Evaluate this ticker IN ISOLATION (no portfolio sizing, no portfolio impact, no correlation talk).
|
|
- Base claims on the provided reports and debate arguments (avoid inventing external macro narratives).
|
|
- Output must be either BUY (go long) or SELL (go short/avoid). If the edge is unclear, pick the less-bad side and set conviction to Low.
|
|
|
|
## DECISION FRAMEWORK (Simple)
|
|
Score each direction 0-10 based on evidence quality and tradeability in the next 5-14 days:
|
|
- Long Edge Score (0-10)
|
|
- Short Edge Score (0-10)
|
|
|
|
Choose the direction with the higher score. If tied, choose BUY.
|
|
|
|
## OUTPUT STRUCTURE (MANDATORY)
|
|
|
|
### Decision
|
|
**DECISION: BUY** or **SELL** (choose exactly one)
|
|
**Conviction: High / Medium / Low**
|
|
**Time Horizon: [X] days**
|
|
|
|
### Trade Setup (Specific)
|
|
- Entry: [price/condition]
|
|
- Stop: [price] ([%] risk)
|
|
- Target: [price] ([%] reward)
|
|
- Risk/Reward: [ratio]
|
|
- Invalidation: [what would prove you wrong]
|
|
- Catalyst / Timing: [next 1-2 weeks drivers]
|
|
|
|
### Why This Should Work
|
|
- [3 bullets max: data-backed reasons]
|
|
|
|
### What Could Break It
|
|
- [2 bullets max: key risks]
|
|
""" + (f"""
|
|
## PAST LESSONS
|
|
Here are reflections on past mistakes - apply these lessons:
|
|
{past_memory_str}
|
|
|
|
**Learning Check:** How are you adjusting based on these past situations?
|
|
""" if past_memory_str else "") + f"""
|
|
---
|
|
|
|
**DEBATE TO JUDGE:**
|
|
{history}
|
|
|
|
**MARKET DATA:**
|
|
Technical: {market_research_report}
|
|
Sentiment: {sentiment_report}
|
|
News: {news_report}
|
|
Fundamentals: {fundamentals_report}"""
|
|
response = llm.invoke(prompt)
|
|
|
|
new_investment_debate_state = {
|
|
"judge_decision": response.content,
|
|
"history": investment_debate_state.get("history", ""),
|
|
"bear_history": investment_debate_state.get("bear_history", ""),
|
|
"bull_history": investment_debate_state.get("bull_history", ""),
|
|
"current_response": response.content,
|
|
"count": investment_debate_state["count"],
|
|
}
|
|
|
|
return {
|
|
"investment_debate_state": new_investment_debate_state,
|
|
"investment_plan": response.content,
|
|
}
|
|
|
|
return research_manager_node
|