Update settings.py

settings.py

@@ -1,49 +1,57 @@
 import os
-from typing import Dict, Any
+from typing import Dict, Any, List
 
+# Residency / deployment guidance (enforce in your infra)
 CANADA_RESIDENCY_REQUIRED = True
 CLOUD_REGION = os.getenv("CLOUD_REGION", "ca-central-1")
 
+# Keys
 COHERE_API_KEY = os.getenv("COHERE_API_KEY", "")
-OPENAI_API_KEY = os.getenv("OPENAI_API_KEY", "")
-ANTHROPIC_API_KEY = os.getenv("ANTHROPIC_API_KEY", "")
 
+# Cohere models (ask Cohere for larger/custom healthcare)
 COHERE_MODEL_PRIMARY = os.getenv("COHERE_MODEL_PRIMARY", "command-r")
 COHERE_EMBED_MODEL = os.getenv("COHERE_EMBED_MODEL", "embed-english-v3.0")
 
-OPENAI_MODEL_FALLBACK = os.getenv("OPENAI_MODEL_FALLBACK", "gpt-4o-mini")
-ANTHROPIC_MODEL_FALLBACK = os.getenv("ANTHROPIC_MODEL_FALLBACK", "claude-3-5-sonnet-latest")
+# Open-model fallbacks (largest first; will try in order)
+OPEN_LLM_CANDIDATES: List[str] = [
+    "Qwen/Qwen2.5-72B-Instruct",           # large & open weights
+    "meta-llama/Meta-Llama-3.1-70B-Instruct",  # large (requires strong GPUs)
+    "mistralai/Mixtral-8x22B-Instruct-v0.1",   # MoE, open
+    "NousResearch/Meta-Llama-3.1-8B-Instruct", # smaller safety net
+    "mistralai/Mistral-7B-Instruct-v0.3"
+]
 
-LOCAL_MODEL_ID = os.getenv("LOCAL_MODEL_ID", "meta-llama/Meta-Llama-3.1-8B-Instruct")
-LOCAL_ENABLE = os.getenv("LOCAL_ENABLE", "0") not in ("0", "false", "False")
-
-USE_SCENARIO_ENGINE = True
+LOCAL_MAX_NEW_TOKENS = int(os.getenv("LOCAL_MAX_NEW_TOKENS", "1200"))
 
+# App knobs
 MODEL_SETTINGS: Dict[str, Any] = {
-    "temperature": float(os.getenv("TEMP", "0.2")),
+    "temperature": float(os.getenv("TEMP", "0.3")),
     "top_p": float(os.getenv("TOP_P", "0.9")),
-    "repetition_penalty": float(os.getenv("REP_PEN", "1.1")),
+    "repetition_penalty": float(os.getenv("REP_PEN", "1.12")),
     "max_new_tokens": int(os.getenv("MAX_NEW_TOKENS", "1500")),
 }
 
 HEALTHCARE_SETTINGS = {
     "supported_file_types": [".csv", ".txt", ".md", ".pdf"],
     "healthcare_keywords": [
-        "hospital", "clinic", "surgery", "wait time", "consult",
-        "beds", "icu", "zone", "health authority", "triage",
-    ],
+        "hospital", "clinic", "surgery", "surgical", "wait time", "consult", "triage",
+        "beds", "occupancy", "icu", "ambulatory", "perioperative", "zone", "health authority",
+        "province", "nova scotia", "iwk"
+    ]
 }
 
-GENERAL_CONVERSATION_PROMPT = "You are a helpful assistant. Be concise, accurate, and friendly."
-
+GENERAL_CONVERSATION_PROMPT = "You are a helpful, concise assistant."
 HEALTHCARE_SYSTEM_PROMPT = (
-    "You are a Canadian healthcare operations analysis copilot. "
-    "Read the entire scenario. Output only a structured JSON plan of tasks. "
-    "Each task must define dataset, filters, grouping, aggregations, joins, pivots, "
-    "and formatting exactly as the scenario requires. "
-    "Do not invent numbers; computations will be executed deterministically."
+    "You are a Canadian healthcare operations copilot. "
+    "You must follow the scenario tasks exactly; calculations are done deterministically. "
+    "Do not invent numbers."
 )
 
+# Feature flags
+USE_SCENARIO_ENGINE = os.getenv("USE_SCENARIO_ENGINE", "1") not in ("0", "false", "False")
+
+# Storage
 DATA_DIR = os.getenv("DATA_DIR", "./data")
 RAG_INDEX_DIR = os.getenv("RAG_INDEX_DIR", "./rag_index")
+PERSIST_CONTENT = False
 SNAPSHOT_PATH = os.getenv("SNAPSHOT_PATH", "./snapshots")