feat: integrate local HF fallback models and enhance Streamlit UI

- Adds support for a local HuggingFace Qwen model (0.5B-Instruct) when OPENAI_API_KEY is not provided
- Adds support for local HuggingFace embeddings (sentence-transformers)
- Enhances Streamlit UI with a minimalist, modern dark theme
- Updates requirements to include transformers, pytorch, and HF dependencies

Co-authored-by: archc0der <119496494+archc0der@users.noreply.github.com>

agent/nodes.py

@@ -5,8 +5,27 @@ from langchain_core.prompts import ChatPromptTemplate
 from agent.state import AgentState
 from rag.retriever import retrieve_documents
 from tools.lead_capture import mock_lead_capture
+from langchain_huggingface import HuggingFacePipeline
+from transformers import pipeline
+import os
+
+_local_llm = None
 
 def get_llm():
+    global _local_llm
+    if not os.environ.get("OPENAI_API_KEY"):
+        if _local_llm is None:
+
+            pipe = pipeline(
+                "text-generation",
+                model="Qwen/Qwen2.5-0.5B-Instruct",
+                max_new_tokens=512,
+                device="cpu",
+                trust_remote_code=True,
+                return_full_text=False
+            )
+            _local_llm = HuggingFacePipeline(pipeline=pipe)
+        return _local_llm
     return ChatOpenAI(model="gpt-4o-mini", temperature=0)
 
 class IntentResponse(BaseModel):
@@ -25,12 +44,26 @@ def detect_intent(state: AgentState) -> AgentState:
         ("user", "{message}")
     ])
 
-    chain = prompt | llm.with_structured_output(IntentResponse)
+    if hasattr(llm, "with_structured_output"):
+        chain = prompt | llm.with_structured_output(IntentResponse)
+    else:
+        from langchain.output_parsers import PydanticOutputParser
+        parser = PydanticOutputParser(pydantic_object=IntentResponse)
+
+        prompt = ChatPromptTemplate.from_messages([
+            ("system", "You are an intent classification assistant for AutoStream. Analyze the user's message and determine the intent. Categories: GREETING, PRODUCT_QUERY, PRICING_QUERY, HIGH_INTENT_LEAD, UNKNOWN. A 'HIGH_INTENT_LEAD' is when a user explicitly expresses interest in signing up, buying, or trying out a plan.\n\n{format_instructions}"),
+            ("user", "{message}")
+        ])
+
+        chain = prompt | llm | parser
 
     history_str = "\n".join([f"{msg['role']}: {msg['content']}" for msg in state.get("conversation_history", [])[-3:]])
     context_message = f"Recent history:\n{history_str}\n\nCurrent message:\n{state['current_message']}"
 
-    response = chain.invoke({"message": context_message})
+    if hasattr(llm, "with_structured_output"):
+        response = chain.invoke({"message": context_message})
+    else:
+        response = chain.invoke({"message": context_message, "format_instructions": parser.get_format_instructions()})
 
     return {"detected_intent": response.intent}
 
@@ -59,7 +92,9 @@ def generate_rag_response(state: AgentState) -> AgentState:
         "message": state["current_message"]
     })
 
-    return {"response": response.content}
+    content = response.content if hasattr(response, "content") else str(response)
+
+    return {"response": content}
 
 def process_lead(state: AgentState) -> AgentState:
     llm = get_llm()
@@ -68,12 +103,25 @@ def process_lead(state: AgentState) -> AgentState:
         ("system", "Extract the user's name, email, and creator platform (e.g. YouTube, TikTok, Instagram) from the message if present. Return null for fields not found."),
         ("user", "{message}")
     ])
-    extract_chain = extract_prompt | llm.with_structured_output(LeadExtractionResponse)
+
+    if hasattr(llm, "with_structured_output"):
+        extract_chain = extract_prompt | llm.with_structured_output(LeadExtractionResponse)
+    else:
+        from langchain.output_parsers import PydanticOutputParser
+        parser = PydanticOutputParser(pydantic_object=LeadExtractionResponse)
+        extract_prompt = ChatPromptTemplate.from_messages([
+            ("system", "Extract the user's name, email, and creator platform (e.g. YouTube, TikTok, Instagram) from the message if present. Return null for fields not found.\n\n{format_instructions}"),
+            ("user", "{message}")
+        ])
+        extract_chain = extract_prompt | llm | parser
 
     history_str = "\n".join([f"{msg['role']}: {msg['content']}" for msg in state.get("conversation_history", [])[-3:]])
     context_message = f"Recent history:\n{history_str}\n\nCurrent message:\n{state['current_message']}"
 
-    extracted = extract_chain.invoke({"message": context_message})
+    if hasattr(llm, "with_structured_output"):
+        extracted = extract_chain.invoke({"message": context_message})
+    else:
+        extracted = extract_chain.invoke({"message": context_message, "format_instructions": parser.get_format_instructions()})
 
     updates = {}
     if extracted.user_name and not state.get("user_name"):

app.py

@@ -4,22 +4,38 @@ from dotenv import load_dotenv
 from agent.graph import app
 from agent.state import AgentState
 
-
 load_dotenv()
 
+st.set_page_config(page_title="AutoStream AI Sales Assistant", page_icon="🎬", layout="centered")
 
-st.set_page_config(page_title="AutoStream AI Sales Assistant", page_icon="🤖", layout="centered")
-
-
+# Custom CSS for minimalist, cooler UI
 st.markdown("""
 <style>
     .stChatFloatingInputContainer {
         bottom: 20px;
     }
+    .main {
+        background-color: #0E1117;
+    }
+    h1 {
+        color: #E2E8F0;
+        font-family: 'Helvetica Neue', Helvetica, Arial, sans-serif;
+        font-weight: 700;
+        text-align: center;
+        margin-bottom: 2rem;
+    }
+    .subtitle {
+        color: #94A3B8;
+        text-align: center;
+        margin-bottom: 2rem;
+        font-size: 1.1rem;
+    }
+    .stAlert {
+        border-radius: 8px;
+    }
 </style>
 """, unsafe_allow_html=True)
 
-
 if "messages" not in st.session_state:
     st.session_state.messages = []
 
@@ -35,54 +51,43 @@ if "messages" not in st.session_state:
         response=""
     )
 
-
     st.session_state.messages.append({"role": "assistant", "content": "Hello! I'm the AutoStream assistant. I can answer questions about our features and pricing. How can I help you today?"})
 
+st.markdown("<h1>🎬 AutoStream Assistant</h1>", unsafe_allow_html=True)
+st.markdown("<div class='subtitle'>Ask about features and pricing, or sign up for a plan instantly!</div>", unsafe_allow_html=True)
 
 if not os.environ.get("OPENAI_API_KEY"):
-    st.warning("⚠️ OPENAI_API_KEY is not set. Please set it in your environment to use the agent.")
-
-st.title("🤖 AutoStream AI Sales Assistant")
-st.markdown("Ask me about AutoStream features and pricing, or sign up for a plan!")
-
+    st.info("ℹ️ OPENAI_API_KEY is not set. The system will fall back to a local Qwen model and HuggingFace embeddings.")
 
 for message in st.session_state.messages:
     with st.chat_message(message["role"]):
         st.markdown(message["content"])
 
-
 if prompt := st.chat_input("What would you like to know?"):
 
     st.session_state.messages.append({"role": "user", "content": prompt})
     with st.chat_message("user"):
         st.markdown(prompt)
 
-
     st.session_state.agent_state["current_message"] = prompt
 
-
     with st.chat_message("assistant"):
         with st.spinner("Thinking..."):
             try:
-
                 result_state = app.invoke(st.session_state.agent_state)
                 st.session_state.agent_state = result_state
 
                 response = result_state["response"]
 
-
                 st.session_state.agent_state["conversation_history"].append({"role": "user", "content": prompt})
                 st.session_state.agent_state["conversation_history"].append({"role": "assistant", "content": response})
 
-
                 if len(st.session_state.agent_state["conversation_history"]) > 12:
                     st.session_state.agent_state["conversation_history"] = st.session_state.agent_state["conversation_history"][-12:]
 
-
                 st.markdown(response)
 
-
-                with st.expander("Agent Reasoning & State"):
+                with st.expander("Agent Reasoning & State", expanded=False):
                     st.write(f"**Detected Intent:** `{result_state.get('detected_intent', 'UNKNOWN')}`")
                     if result_state.get("retrieved_documents") and result_state.get("detected_intent") in ["PRODUCT_QUERY", "PRICING_QUERY"]:
                         st.write(f"**RAG Retrieval:** Found {len(result_state['retrieved_documents'])} relevant knowledge chunks.")
@@ -99,5 +104,4 @@ if prompt := st.chat_input("What would you like to know?"):
                 response = f"An error occurred: {str(e)}"
                 st.error(response)
 
-
     st.session_state.messages.append({"role": "assistant", "content": response})

rag/embeddings.py

@@ -1,7 +1,9 @@
 from langchain_openai import OpenAIEmbeddings
+from langchain_huggingface import HuggingFaceEmbeddings
+import os
 
 def get_embeddings():
-    """
-    Returns the embedding model used for the RAG pipeline.
-    """
+
+    if not os.environ.get("OPENAI_API_KEY"):
+        return HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
     return OpenAIEmbeddings()

requirements.txt

@@ -9,3 +9,8 @@ pydantic
 pytest
 pytest-mock
 streamlit
+transformers
+langchain-huggingface
+huggingface-hub
+sentence-transformers
+torch