change external source recommener

src/streamlit_app.py

@@ -256,7 +256,7 @@ Guidelines:
 - If the subtask implies a taxonomic or common name group (e.g., frog, snake, salmon), apply CONTAINS or STARTS WITH filters on Species.name or species_full_name, using toLower(...) for case-insensitive matching.
 - If the subtask includes a time range, include date filtering.
 - Prefer using DISTINCT to avoid redundant results.
-- Only return fields that are clearly needed to fulfill the subtask.
+- Only return fields that are clearl y needed to fulfill the subtask.
 
 Return your response strictly as a **JSON object** with the following fields:
 - "intent": a short description of what the query does
@@ -466,17 +466,23 @@ Plain text only — no code fences. Markdown link syntax (`[text](url)`) is allo
 
 
 
-
 def external_resource_recommender(subtask: str, client=openai_client) -> str:
     prompt = f"""
-You are a helpful assistant for researchers. Please recommend 3 reliable and relevant online datasets or websites that can help with the following subtask:
+You are a helpful research assistant. Your task is to recommend **three reliable, publicly accessible online datasets or data repositories** that can assist with the following scientific subtask:
 
-"{subtask}"
+{subtask}
+
+Only include sources that are:
+- Trusted (e.g., government, academic, or well-established platforms)
+- Relevant to the topic
+- Accessible without login when possible
 
-Format your output in markdown as:
+Format your answer strictly in markdown:
 - [Name of Source](URL)
 - [Name of Source](URL)
 - [Name of Source](URL)
+
+Do not include any explanations or extra text—only the list.
 """
     rsp = client.chat.completions.create(                      
         model="gpt-4o",
@@ -487,6 +493,7 @@ Format your output in markdown as:
 
 
 
+
 def fallback_query_router(subtask: str, driver) -> pd.DataFrame:
     text = subtask.lower()