Upload 6 files

model.py

@@ -1,6 +1,6 @@
 # model.py
 from sentence_transformers import SentenceTransformer, util
-from sql_templates import sql_templates  # new import
+from sql_templates import sql_templates, sql_keyword_aliases, fuzzy_aliases
 import torch
 
 # Load training rules (string-to-SQL map)
@@ -16,35 +16,44 @@ train_embeddings = model.encode(train_prompts, convert_to_tensor=True)
 
 def oracle_sql_suggester(prompt):
     prompt_clean = prompt.strip().lower()
-    
+
     # Try direct rule match
     if prompt_clean in rules:
         return rules[prompt_clean]
+
+    # Check template keywords (first!)
+    for word in prompt_clean.split():
+     if word in sql_keyword_aliases:
+        mapped_key = sql_keyword_aliases[word]
+        return sql_templates[mapped_key]
     
-    # Semantic matching
-    user_embedding = model.encode(prompt_clean, convert_to_tensor=True)
-    cosine_scores = util.cos_sim(user_embedding, train_embeddings)
-    
-    top_match_index = torch.argmax(cosine_scores).item()
-    top_score = cosine_scores[0][top_match_index].item()
+    for key, template in sql_templates.items():
+     if key.replace("_", " ") in prompt_clean or key in prompt_clean:
+        return template
+# Try keyword aliases (word-level match)
+    for word in prompt_clean.split():
+     if word in sql_keyword_aliases:
+       return sql_templates[sql_keyword_aliases[word]]  
+   
+# Try fuzzy alias matches
+    for fuzzy_phrase, mapped_key in fuzzy_aliases.items():
+     if fuzzy_phrase in prompt_clean:
+        return sql_templates[mapped_key]
+
+  
     
-    if top_score >= 0.7:
-        matched_prompt = train_prompts[top_match_index]
-        return rules[matched_prompt]
     
-     # Check template keywords
-    for key in sql_templates:
-        if key.replace("_", " ") in prompt_clean or key in prompt_clean:
-            return sql_templates[key]
-
-    # Semantic match
+    # Semantic matching
     user_embedding = model.encode(prompt_clean, convert_to_tensor=True)
     cosine_scores = util.cos_sim(user_embedding, train_embeddings)
+
     top_match_index = torch.argmax(cosine_scores).item()
     top_score = cosine_scores[0][top_match_index].item()
 
     if top_score >= 0.7:
         matched_prompt = train_prompts[top_match_index]
         return rules[matched_prompt]
-    
-    return "🤖 Sorry, I couldn’t understand that. Please try rephrasing your request."
+
+    # Fallback
+    return "🤖 Sorry, I couldn’t understand that. Please try rephrasing your request."    
+    

sql_templates.py

@@ -1,3 +1,5 @@
+from collections import defaultdict
+
 sql_templates = {
     "basic_select": "SELECT column1, column2 FROM table_name;",
     "select_where": "SELECT column1 FROM table_name WHERE condition;",
@@ -8,3 +10,32 @@ sql_templates = {
     "update": "UPDATE table_name SET col1 = val1 WHERE condition;",
     "delete": "DELETE FROM table_name WHERE condition;"
 }
+
+sql_keyword_aliases = {
+    "select": "basic_select",
+    "where": "select_where",
+    "join": "join_example",
+    "group": "group_by",
+    "group by": "group_by",
+    "having": "having",
+    "insert": "insert",
+    "update": "update",
+    "delete": "delete"
+}
+
+# 🧠 NEW fuzzy aliases
+fuzzy_aliases = {
+    "grouped result": "group_by",
+    "combine tables": "join_example",
+    "add new row": "insert",
+    "modify records": "update",
+    "remove entry": "delete",
+    "get rows": "basic_select",
+    "filter records": "select_where",
+    "apply condition": "select_where",
+    "summarize": "group_by",
+    "count groups": "group_by",
+    "condition on groups": "having",
+    "change row": "update",
+    "erase record": "delete"
+}