updates

agents/intent_classifier.py

@@ -88,10 +88,12 @@ _PRONOUN_ONLY_PATTERN = re.compile(
 )
 
 _WORKED_EXAMPLE_PATTERN = re.compile(
-    r"\d+(?:\s*,\s*\d+){1,}.*\b(this|that|same)\b.*\b(search|sort|algorithm|method|approach|traversal)\b",
+    r"\d+(?:\s*,\s*\d+){1,}.*\b(this|that|same|additional info|additional information)\b.*\b(search|sort|algorithm|method|approach|traversal)\b",
     re.IGNORECASE,
 )
 
+_CODE_REQUEST_PATTERN = re.compile(r"\b(code|implementation|logic)\b", re.IGNORECASE)
+
 def is_followup_query(query: str, has_conversation_history: bool) -> bool:
     """
     A query is treated as a follow-up when:
@@ -110,10 +112,12 @@ def is_followup_query(query: str, has_conversation_history: bool) -> bool:
         return True
 
     word_count = len(query.strip().split())
-    has_pronoun = bool(_PRONOUN_ONLY_PATTERN.search(query))
+    has_pronoun = bool(_PRONOUN_ONLY_PATTERN.search(query)) or bool(_CODE_REQUEST_PATTERN.search(query))
     mentions_topic = bool(extract_topics(query))
 
-    if has_pronoun and not mentions_topic and word_count <= 8:
+    # if has_pronoun and not mentions_topic and word_count <= 8:
+    #     return True
+    if not mentions_topic and word_count <= 8 and has_conversation_history:
         return True
 
     return False

config.py

@@ -77,6 +77,7 @@ LOCAL_MODEL_TEMPERATURE = float(os.getenv("LOCAL_MODEL_TEMPERATURE", 0.3))
 LOCAL_MODEL_LOAD_IN_4BIT = os.getenv("LOCAL_MODEL_LOAD_IN_4BIT", "True") == "True"
 
 GROQ_API_KEY = os.getenv("GROQ_API_KEY")
+GROQ_MAX_TOKENS = int(os.getenv("GROQ_MAX_TOKENS", 1024))
 GROQ_MODEL_NAME = os.getenv("GROQ_MODEL_NAME", "llama-3.3-70b-versatile")
 
 OPENAI_API_KEY = os.getenv("OPENAI_API_KEY", "")

llm/groq_provider.py

@@ -47,7 +47,7 @@ class GroqProvider(LLMProvider):
                 model=self._model,
                 messages=messages,
                 temperature=config.LOCAL_MODEL_TEMPERATURE,
-                max_tokens=config.LOCAL_MODEL_MAX_NEW_TOKENS,
+                max_tokens=config.GROQ_MAX_TOKENS,
                 stream=False,
             )
         except Exception:

llm/prompts.py

@@ -24,7 +24,10 @@ These builders only assemble strings — they never call the LLM themselves
 
 SYSTEM_PROMPT = (
     "You are a patient, precise Data Structures and Algorithms (DSA) tutor. "
-    "Answer using ONLY the information provided in the retrieved context below. "
+    "Answer using the retrieved context as your primary source of truth for explanations, complexity, and examples."
+    "If the user asks for code, you may write a clean, correct implementation in the requested language, consistent with the algorithm described in the context."
+    "If language not specified, use python as the primary language."
+    "Always include the time and space complexities wherever possible."
     "If the retrieved context does not contain enough information to answer "
     "confidently, say so explicitly rather than guessing or inventing details. "
     "Explain concepts clearly, use the time/space complexity and examples given "

requirements.txt

@@ -7,7 +7,7 @@ chromadb>=1.0.0
 sentence-transformers==3.3.1
 mem0ai==0.1.29
 groq==0.15.0
-torch==2.8.0
+torch==2.6.0
 transformers==4.47.1
 accelerate==1.2.1
 bitsandbytes==0.45.0