Upload app.py with huggingface_hub

app.py

@@ -0,0 +1,195 @@
+import streamlit as st
+import json
+import os
+from datetime import datetime
+from typing import Dict, List, Any
+
+# ============== LLM / RAG deps ==============
+from langchain_openai import ChatOpenAI
+from langchain_community.vectorstores import Chroma
+from langchain_community.embeddings.sentence_transformer import SentenceTransformerEmbeddings
+from langchain_core.prompts import ChatPromptTemplate
+from langchain_core.tools import tool
+from langchain.agents import create_tool_calling_agent, AgentExecutor
+
+# Optional: Llama Guard (Groq)
+try:
+    from groq import Groq
+    HAS_GROQ = True
+except Exception:
+    HAS_GROQ = False
+
+# ============== CONFIG ==============
+# TODO: set via env or secrets in Streamlit Cloud
+API_KEY = os.getenv("OPENAI_API_KEY", "gl-U2FsdGVkX1+r0wSt3dbixZ6yKDLw0Rg46XrTm+rJY/t9b+4TU3aqZ4eDbA2OHufX")
+API_BASE = os.getenv("OPENAI_API_BASE", "https://api.openai.com/v1")  # or your compatible gateway
+MODEL_NAME = os.getenv("OPENAI_MODEL", "gpt-4o-mini")
+
+# Your nutrition vectorstore (text hypotheticals)
+# TODO: update if you used different names/locations
+PERSIST_DIR = os.getenv("CHROMA_DIR", "./research_db_hypotheticals")
+COLLECTION_NAME = os.getenv("CHROMA_COLLECTION", "hypothetical_questions_text")
+
+# Optional Llama Guard key
+GROQ_API_KEY = os.getenv("LLAMA_API_KEY", "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")
+
+# ============== LOAD VECTORSTORE ==============
+# If you embedded with OpenAI, you can switch to an OpenAIEmbeddings. If you used gte/gte-large (HF), match that here.
+# Using a robust open model by default.
+embedding_model = SentenceTransformerEmbeddings(model_name="thenlper/gte-large")
+
+vector_store = Chroma(
+    collection_name=COLLECTION_NAME,
+    persist_directory=PERSIST_DIR,
+    embedding_function=embedding_model
+)
+
+retriever = vector_store.as_retriever(
+    search_type="similarity",
+    search_kwargs={"k": 5}
+)
+
+# ============== LLM CLIENT ==============
+llm = ChatOpenAI(
+    model=MODEL_NAME,
+    temperature=0,
+    max_retries=3,
+    api_key=API_KEY,
+    base_url=API_BASE,
+)
+
+# ============== SAFETY FILTER (Optional) ==============
+def filter_input_with_llama_guard(user_input: str, model: str = "meta-llama/llama-guard-4-12b") -> str:
+    """
+    Use Llama Guard (Groq) to sanitize user input. Returns the moderated text if available,
+    else returns the original text.
+    """
+    if not HAS_GROQ or not GROQ_API_KEY:
+        return user_input  # no safety client configured
+
+    try:
+        client = Groq(api_key=GROQ_API_KEY)
+        resp = client.chat.completions.create(
+            model=model,
+            messages=[
+                {"role": "system", "content": "You are a safety filter. Return only sanitized or safe text."},
+                {"role": "user", "content": user_input},
+            ],
+        )
+        safe_text = resp.choices[0].message.content.strip()
+        return safe_text or user_input
+    except Exception:
+        return user_input
+
+# ============== SIMPLE RAG PIPELINE ==============
+RAG_SYSTEM_MESSAGE = """You are a medical-support assistant specializing in nutritional disorders.
+Answer clearly, concisely, and factually. Use ONLY the provided context.
+If the answer is not contained in the context, say: "I don't know." 
+Prefer listing symptoms, diagnosis criteria, risk factors, and treatments/dietary recommendations when relevant.
+Avoid speculation; cite nothing to the user."""
+
+RAG_USER_TEMPLATE = """###Context
+{context}
+
+###Question
+{question}
+"""
+
+rag_prompt = ChatPromptTemplate.from_messages([
+    ("system", RAG_SYSTEM_MESSAGE),
+    ("user", RAG_USER_TEMPLATE),
+])
+
+@tool("agentic_rag")
+def agentic_rag(question: str) -> str:
+    """
+    Nutrition Disorder RAG: retrieves context from Chroma and answers strictly from it.
+    """
+    try:
+        docs = retriever.invoke(question)
+        if not docs:
+            return "I don't know."
+
+        context = "\n\n".join([d.page_content for d in docs])
+        chain = rag_prompt | llm
+        resp = chain.invoke({"context": context, "question": question})
+        return resp.content if hasattr(resp, "content") else str(resp)
+    except Exception as e:
+        return f"Error: {e}"
+
+# ============== STREAMLIT UI ==============
+st.set_page_config(page_title="Nutrition Disorder Agentic RAG", page_icon="🥦", layout="centered")
+
+def ensure_state():
+    if "logged_in" not in st.session_state:
+        st.session_state.logged_in = False
+    if "user_name" not in st.session_state:
+        st.session_state.user_name = ""
+    if "history" not in st.session_state:
+        st.session_state.history = [{"role": "assistant", "content": "Welcome! I’m your Nutrition Disorder assistant. How can I help?"}]
+
+def login_page():
+    st.title("Nutrition Disorder Agent — Login")
+    with st.form("login_form"):
+        name = st.text_input("Your name")
+        submitted = st.form_submit_button("Enter")
+    if submitted:
+        if name.strip():
+            st.session_state.logged_in = True
+            st.session_state.user_name = name.strip()
+            st.success(f"Welcome, {name.strip()}!")
+            st.rerun()
+        else:
+            st.error("Please enter your name.")
+
+def chat_page():
+    st.title("Nutrition Disorder Agent")
+    st.caption("Evidence-grounded answers about symptoms, diagnosis, and treatments for nutritional disorders.")
+
+    # Show chat history
+    for m in st.session_state.history:
+        with st.chat_message(m["role"]):
+            st.markdown(m["content"])
+
+    # Input
+    user_msg = st.chat_input("Ask about symptoms, diagnosis, or treatments…")
+    if user_msg:
+        # Optional safety filter
+        filtered = filter_input_with_llama_guard(user_msg)
+
+        # Append user
+        st.session_state.history.append({"role": "user", "content": user_msg})
+        with st.chat_message("user"):
+            st.markdown(user_msg)
+
+        # Call tool
+        with st.spinner("Thinking..."):
+            try:
+                answer = agentic_rag.invoke(filtered)
+            except Exception as e:
+                answer = f"Sorry, I hit an error: {e}"
+
+        st.session_state.history.append({"role": "assistant", "content": answer})
+        with st.chat_message("assistant"):
+            st.markdown(answer)
+
+    # Footer controls
+    cols = st.columns(3)
+    if cols[0].button("Reset Chat"):
+        st.session_state.history = [{"role": "assistant", "content": "Welcome! I’m your Nutrition Disorder assistant. How can I help?"}]
+        st.rerun()
+    with cols[1]:
+        st.write("")
+    if cols[2].button("Logout"):
+        st.session_state.clear()
+        st.rerun()
+
+def main():
+    ensure_state()
+    if not st.session_state.logged_in:
+        login_page()
+    else:
+        chat_page()
+
+if __name__ == "__main__":
+    main()