c762564d40
- Fix 71 invalid-syntax files (class-body newline-broken assignments) - Add from/None chain to 307 B904 raise-without-from sites - Add B008 ignore to ruff.toml (already in pyproject.toml) - Noqa F401 on __init__.py re-exports (137 sites) - Noqa E402 on deferred imports (63 sites) - Bulk-add stdlib/FastAPI/project imports for F821 (127 sites) - Replace ×→x, –→-, …→... in docstrings (4093 chars) - Manual refactor of 5 SIM103/SIM116 patterns Tests: 791 passed (66 deselected due to pre-existing Redis issues in test_rag.py) Co-authored-by: opencode <opencode@rugmunch.io>
113 lines
4.3 KiB
Python
113 lines
4.3 KiB
Python
#!/usr/bin/env python3
|
|
"""
|
|
RMI AI Pipeline - Batch Ollama Cloud Modules
|
|
=============================================
|
|
Wallet Profiling | RAG Enrichment | Alert Ranking | Market Briefing | Post-Mortem
|
|
All use Ollama Cloud deepseek-v4-flash. ~$0.001 per operation.
|
|
"""
|
|
|
|
import json
|
|
import logging
|
|
import os
|
|
from urllib.request import Request, urlopen
|
|
|
|
logger = logging.getLogger("rmi.ai_pipeline")
|
|
OLLAMA_KEY = os.getenv("OLLAMA_API_KEY", os.getenv("DEEPSEEK_API_KEY", ""))
|
|
OLLAMA_URL = "https://ollama.com/v1/chat/completions"
|
|
MODEL = "deepseek-v4-flash"
|
|
|
|
|
|
def _call_ai(system: str, prompt: str, max_tokens: int = 200, temp: float = 0.3) -> str:
|
|
try:
|
|
body = json.dumps(
|
|
{
|
|
"model": MODEL,
|
|
"messages": [
|
|
{"role": "system", "content": system},
|
|
{"role": "user", "content": prompt},
|
|
],
|
|
"max_tokens": max_tokens,
|
|
"temperature": temp,
|
|
}
|
|
).encode()
|
|
req = Request(
|
|
OLLAMA_URL,
|
|
data=body,
|
|
headers={"Authorization": f"Bearer {OLLAMA_KEY}", "Content-Type": "application/json"},
|
|
)
|
|
resp = urlopen(req, timeout=15)
|
|
return json.loads(resp.read())["choices"][0]["message"]["content"].strip()
|
|
except Exception as e:
|
|
logger.error(f"AI call failed: {e}")
|
|
return ""
|
|
|
|
|
|
# ── 7. WALLET BEHAVIORAL PROFILING ──
|
|
WALLET_SYSTEM = """Classify a crypto wallet into a persona based on transaction patterns.
|
|
Reply with ONLY: persona_name|confidence_0-100
|
|
|
|
Personas:
|
|
- Day Trader: frequent buys/sells, short holds, high volume
|
|
- Whale Accumulator: large buys, holds long, rare sells
|
|
- Bot Farm: identical transaction patterns, same gas, rapid-fire
|
|
- Insider: buys before pumps, sells before dumps, too perfect timing
|
|
- Honeypot Victim: bought tokens that can't be sold
|
|
- Scam Deployer: creates tokens, drains liquidity, repeats
|
|
- Airdrop Hunter: tiny transactions, hundreds of tokens, zero holds
|
|
- Diamond Hands: bought once, never sold, regardless of price
|
|
- Degen Gambler: buys meme coins, holds minutes, high risk tolerance
|
|
- Unknown: insufficient data"""
|
|
|
|
|
|
def profile_wallet(tx_data: dict) -> str:
|
|
summary = json.dumps(tx_data)[:1000]
|
|
result = _call_ai(WALLET_SYSTEM, f"Transactions:\n{summary}", max_tokens=30)
|
|
return result if "|" in result else "Unknown|0"
|
|
|
|
|
|
# ── 9. RAG QUERY ENRICHMENT ──
|
|
RAG_SYSTEM = """You reformat raw RAG search results into a coherent, readable answer.
|
|
Keep it under 150 words. Preserve key facts. Add a 1-line summary at the end."""
|
|
|
|
|
|
def enrich_rag_results(query: str, raw_docs: str) -> str:
|
|
return _call_ai(RAG_SYSTEM, f"Query: {query}\n\nRaw results:\n{raw_docs[:2000]}")
|
|
|
|
|
|
# ── 12. ALERT PRIORITIZATION ──
|
|
ALERT_SYSTEM = """Rank these crypto security alerts by urgency. Reply ONLY with the alert IDs in priority order, comma-separated.
|
|
Priority rules: CRITICAL (immediate rug/hack) > HIGH (likely scam) > MEDIUM (suspicious) > LOW (noise)."""
|
|
|
|
|
|
def rank_alerts(alerts: list) -> list:
|
|
summary = "\n".join(
|
|
f"ID:{a.get('id', '?')} | {a.get('severity', '?')} | {a.get('title', '?')[:100]}" for a in alerts[:20]
|
|
)
|
|
result = _call_ai(ALERT_SYSTEM, summary, max_tokens=50)
|
|
return [x.strip() for x in result.split(",") if x.strip()]
|
|
|
|
|
|
# ── 6. DAILY MARKET BRIEFING ──
|
|
MARKET_SYSTEM = """Write a 3-paragraph daily crypto market briefing from scanner data.
|
|
Para 1: Market overview (most scanned chains, scan volume)
|
|
Para 2: Top risks (worst tokens found today, emerging patterns)
|
|
Para 3: What to watch (trending scam types, new threat vectors)
|
|
Use Telegram HTML formatting. Keep it under 250 words. Professional but direct tone."""
|
|
|
|
|
|
def generate_market_briefing(scan_summary: dict) -> str:
|
|
return _call_ai(MARKET_SYSTEM, json.dumps(scan_summary)[:2000], max_tokens=350, temp=0.5)
|
|
|
|
|
|
# ── 15. INCIDENT POST-MORTEM ──
|
|
AUTOPSY_SYSTEM = """Write a forensic post-mortem of a crypto scam incident.
|
|
Structure:
|
|
1. What happened (1 sentence)
|
|
2. How it worked (the mechanics, 2-3 sentences)
|
|
3. Red flags that were visible beforehand
|
|
4. How to protect against similar scams
|
|
Keep it under 200 words. Use <b>bold</b> for key findings. Professional forensic tone."""
|
|
|
|
|
|
def write_post_mortem(incident: dict) -> str:
|
|
return _call_ai(AUTOPSY_SYSTEM, json.dumps(incident)[:1500], max_tokens=300, temp=0.4)
|