Update app.py

app.py

@@ -1,275 +1,243 @@
-# import os
-# import sys
-# import requests
-# from langchain.chains import ConversationalRetrievalChain
-# from langchain.document_loaders import PyPDFLoader, Docx2txtLoader, TextLoader
-# from langchain.text_splitter import CharacterTextSplitter
-# from langchain.vectorstores import Chroma
-# from langchain.embeddings import HuggingFaceEmbeddings
-# from langchain.llms.base import LLM
-# import gradio as gr
-
-# # workaround for sqlite in HF spaces
-# __import__('pysqlite3')
-# sys.modules['sqlite3'] = sys.modules.pop('pysqlite3')
-
-# # 📄 Load documents
-# docs = []
-# for f in os.listdir("multiple_docs"):
-#     if f.endswith(".pdf"):
-#         loader = PyPDFLoader(os.path.join("multiple_docs", f))
-#         docs.extend(loader.load())
-#     elif f.endswith(".docx") or f.endswith(".doc"):
-#         loader = Docx2txtLoader(os.path.join("multiple_docs", f))
-#         docs.extend(loader.load())
-#     elif f.endswith(".txt"):
-#         loader = TextLoader(os.path.join("multiple_docs", f))
-#         docs.extend(loader.load())
-
-# # 🔗 Split into chunks
-# splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=10)
-# docs = splitter.split_documents(docs)
-
-# texts = [doc.page_content for doc in docs]
-# metadatas = [{"id": i} for i in range(len(texts))]
-
-# # 🧠 Embeddings
-# embedding_function = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
-
-# # 🗃️ Vectorstore
-# vectorstore = Chroma(
-#     persist_directory="./db",
-#     embedding_function=embedding_function
-# )
-# vectorstore.add_texts(texts=texts, metadatas=metadatas)
-# vectorstore.persist()
-
-# # 🔐 Get DeepSeek API key from env
-# DEEPSEEK_API_KEY = os.getenv("DEEPSEEK_API_KEY")
-# if DEEPSEEK_API_KEY is None:
-#     raise ValueError("DEEPSEEK_API_KEY environment variable is not set.")
-
-# # 🌟 DeepSeek API endpoint
-# DEEPSEEK_API_URL = "https://api.deepseek.com/v1/chat/completions"
-
-# # 🔷 Wrap DeepSeek API into LangChain LLM
-# class DeepSeekLLM(LLM):
-#     """LLM that queries DeepSeek's API."""
-#     api_key: str = DEEPSEEK_API_KEY
-
-#     def _call(self, prompt, stop=None, run_manager=None, **kwargs):
-#         headers = {
-#             "Authorization": f"Bearer {self.api_key}",
-#             "Content-Type": "application/json"
-#         }
-#         payload = {
-#             "model": "deepseek-chat",   # adjust if you have a specific model name
-#             "messages": [
-#                 {"role": "system", "content": "You are a helpful assistant."},
-#                 {"role": "user", "content": prompt}
-#             ],
-#             "temperature": 0.7,
-#             "max_tokens": 512
-#         }
-#         response = requests.post(DEEPSEEK_API_URL, headers=headers, json=payload)
-#         response.raise_for_status()
-#         data = response.json()
-#         return data["choices"][0]["message"]["content"].strip()
-
-#     @property
-#     def _llm_type(self) -> str:
-#         return "deepseek_api"
-
-# llm = DeepSeekLLM()
-
-# # 🔗 Conversational chain
-# chain = ConversationalRetrievalChain.from_llm(
-#     llm,
-#     retriever=vectorstore.as_retriever(search_kwargs={'k': 6}),
-#     return_source_documents=True,
-#     verbose=False
-# )
-
-# # 💬 Gradio UI
-# chat_history = []
-
-# with gr.Blocks() as demo:
-#     chatbot = gr.Chatbot(
-#         [("", "Hello, I'm Thierry Decae's chatbot, you can ask me any recruitment related questions such as my experience, where I'm eligible to work, skills etc you can chat with me directly in multiple languages")],
-#         avatar_images=["./multiple_docs/Guest.jpg", "./multiple_docs/Thierry Picture.jpg"]
-#     )
-#     msg = gr.Textbox(placeholder="Type your question here...")
-#     clear = gr.Button("Clear")
-
-#     def user(query, chat_history):
-#         chat_history_tuples = [(m[0], m[1]) for m in chat_history]
-#         result = chain({"question": query, "chat_history": chat_history_tuples})
-#         chat_history.append((query, result["answer"]))
-#         return gr.update(value=""), chat_history
-
-#     msg.submit(user, [msg, chatbot], [msg, chatbot], queue=False)
-#     clear.click(lambda: None, None, chatbot, queue=False)
-
-# demo.launch(debug=True)  # remove share=True if running in HF Spaces
-
-
 import os
 import sys
 import requests
-from langchain.chains import ConversationalRetrievalChain, LLMChain
-from langchain.document_loaders import PyPDFLoader, Docx2txtLoader, TextLoader
-from langchain.text_splitter import CharacterTextSplitter
-from langchain.vectorstores import Chroma
-from langchain.embeddings import HuggingFaceEmbeddings
-from langchain.llms.base import LLM
-from langchain.prompts import PromptTemplate
-from langchain.chains.question_answering import load_qa_chain
+
+# SQLite workaround for Chroma on Hugging Face Spaces
+try:
+    __import__("pysqlite3")
+    sys.modules["sqlite3"] = sys.modules.pop("pysqlite3")
+except Exception:
+    pass
+
 import gradio as gr
 
-# workaround for sqlite in HF spaces
-__import__('pysqlite3')
-sys.modules['sqlite3'] = sys.modules.pop('pysqlite3')
-
-# 📄 Load documents
-docs = []
-for f in os.listdir("multiple_docs"):
-    if f.endswith(".pdf"):
-        loader = PyPDFLoader(os.path.join("multiple_docs", f))
-        docs.extend(loader.load())
-    elif f.endswith(".docx") or f.endswith(".doc"):
-        loader = Docx2txtLoader(os.path.join("multiple_docs", f))
-        docs.extend(loader.load())
-    elif f.endswith(".txt"):
-        loader = TextLoader(os.path.join("multiple_docs", f))
-        docs.extend(loader.load())
-
-# 🔗 Split into chunks
-splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=10)
-docs = splitter.split_documents(docs)
-
-texts = [doc.page_content for doc in docs]
-metadatas = [{"id": i} for i in range(len(texts))]
-
-# 🧠 Embeddings
-embedding_function = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
-
-# 🗃️ Vectorstore
-vectorstore = Chroma(
-    persist_directory="./db",
-    embedding_function=embedding_function
-)
-vectorstore.add_texts(texts=texts, metadatas=metadatas)
-vectorstore.persist()
+from langchain_community.document_loaders import PyPDFLoader, Docx2txtLoader, TextLoader
+from langchain_text_splitters import CharacterTextSplitter
+from langchain_chroma import Chroma
+from langchain_huggingface import HuggingFaceEmbeddings
 
-# 🔐 Get DeepSeek API key from env
-DEEPSEEK_API_KEY = os.getenv("DEEPSEEK_API_KEY")
-if DEEPSEEK_API_KEY is None:
-    raise ValueError("DEEPSEEK_API_KEY environment variable is not set.")
 
-# 🌟 DeepSeek API endpoint
+DOCS_DIR = "multiple_docs"
+DB_DIR = "./db"
+COLLECTION_NAME = "thierry_recruitment_docs"
+
+DEEPSEEK_API_KEY = os.getenv("DEEPSEEK_API_KEY")
 DEEPSEEK_API_URL = "https://api.deepseek.com/v1/chat/completions"
 
-# 🔷 Wrap DeepSeek API into LangChain LLM
-class DeepSeekLLM(LLM):
-    """LLM that queries DeepSeek's API."""
-    api_key: str = DEEPSEEK_API_KEY
-
-    def _call(self, prompt, stop=None, run_manager=None, **kwargs):
-        headers = {
-            "Authorization": f"Bearer {self.api_key}",
-            "Content-Type": "application/json"
-        }
-        payload = {
-            "model": "deepseek-chat",   # adjust if you have a specific model name
-            "messages": [
-                {"role": "system", "content": "You are a helpful assistant."},
-                {"role": "user", "content": prompt}
-            ],
-            "temperature": 0.7,
-            "max_tokens": 512
-        }
-        response = requests.post(DEEPSEEK_API_URL, headers=headers, json=payload)
-        response.raise_for_status()
-        data = response.json()
-        return data["choices"][0]["message"]["content"].strip()
-
-    @property
-    def _llm_type(self) -> str:
-        return "deepseek_api"
-
-llm = DeepSeekLLM()
-
-# ✨ Custom prompt template
-template = """
-You are Thierry Decae's chatbot. Your role is to answer questions about his career, experience, availability — in other words
-any recruitment-related question.
-Use the following context to answer the user's question as fully and accurately as possible.
-If you don't know the answer, say "I'm not sure about that."
-Always answer as if you were Thierry Decae — do not refer to him as 'he', use 'I' instead.
-
-Context:
+WELCOME_MESSAGE = (
+    "Hello, I'm Thierry Decae's chatbot. You can ask me recruitment-related "
+    "questions about my experience, skills, availability, work eligibility, "
+    "projects, and background. You can chat with me in multiple languages."
+)
+
+
+def call_deepseek(messages, temperature=0.4, max_tokens=700):
+    if not DEEPSEEK_API_KEY:
+        return "DEEPSEEK_API_KEY is not set in the Hugging Face Space secrets."
+
+    headers = {
+        "Authorization": f"Bearer {DEEPSEEK_API_KEY}",
+        "Content-Type": "application/json",
+    }
+
+    payload = {
+        "model": "deepseek-chat",
+        "messages": messages,
+        "temperature": temperature,
+        "max_tokens": max_tokens,
+    }
+
+    response = requests.post(
+        DEEPSEEK_API_URL,
+        headers=headers,
+        json=payload,
+        timeout=60,
+    )
+    response.raise_for_status()
+    data = response.json()
+    return data["choices"][0]["message"]["content"].strip()
+
+
+def load_documents():
+    if not os.path.exists(DOCS_DIR):
+        raise FileNotFoundError(f"Folder not found: {DOCS_DIR}")
+
+    docs = []
+
+    for filename in os.listdir(DOCS_DIR):
+        path = os.path.join(DOCS_DIR, filename)
+        lower = filename.lower()
+
+        try:
+            if lower.endswith(".pdf"):
+                loader = PyPDFLoader(path)
+                docs.extend(loader.load())
+            elif lower.endswith(".docx") or lower.endswith(".doc"):
+                loader = Docx2txtLoader(path)
+                docs.extend(loader.load())
+            elif lower.endswith(".txt"):
+                loader = TextLoader(path, encoding="utf-8")
+                docs.extend(loader.load())
+        except Exception as e:
+            print(f"Could not load {filename}: {e}", flush=True)
+
+    if not docs:
+        raise ValueError(f"No documents found in {DOCS_DIR}")
+
+    splitter = CharacterTextSplitter(
+        chunk_size=1000,
+        chunk_overlap=100,
+    )
+
+    return splitter.split_documents(docs)
+
+
+def build_vectorstore():
+    print("Loading embedding model...", flush=True)
+
+    embedding_function = HuggingFaceEmbeddings(
+        model_name="sentence-transformers/all-MiniLM-L6-v2"
+    )
+
+    print("Loading documents...", flush=True)
+    docs = load_documents()
+
+    print(f"Loaded {len(docs)} document chunks.", flush=True)
+
+    vectorstore = Chroma.from_documents(
+        documents=docs,
+        embedding=embedding_function,
+        persist_directory=DB_DIR,
+        collection_name=COLLECTION_NAME,
+    )
+
+    return vectorstore
+
+
+vectorstore = build_vectorstore()
+retriever = vectorstore.as_retriever(search_kwargs={"k": 6})
+
+
+def format_chat_history(history):
+    if not history:
+        return ""
+
+    lines = []
+    for message in history[-8:]:
+        role = message.get("role", "")
+        content = message.get("content", "")
+        if role and content:
+            lines.append(f"{role}: {content}")
+
+    return "\n".join(lines)
+
+
+def answer_question(query, chat_history):
+    if not query or not query.strip():
+        return "", chat_history
+
+    query = query.strip()
+
+    try:
+        retrieved_docs = retriever.invoke(query)
+
+        context = "\n\n".join(
+            doc.page_content for doc in retrieved_docs if doc.page_content
+        )
+
+        history_text = format_chat_history(chat_history)
+
+        system_prompt = """
+You are Thierry Decae's recruitment chatbot.
+
+Your role is to answer questions about Thierry Decae's career, experience,
+skills, projects, availability, work eligibility, and professional background.
+
+Use only the provided context when answering factual questions.
+If the answer is not available in the context, say: "I'm not sure about that."
+
+Always answer as Thierry, using "I", "my", and "me".
+Do not refer to Thierry as "he".
+Be professional, concise, and helpful.
+You may answer in the same language as the user.
+"""
+
+        user_prompt = f"""
+Conversation history:
+{history_text}
+
+Context from Thierry's documents:
 {context}
 
-Question: {question}
+User question:
+{query}
 
 Answer:
 """
 
-prompt = PromptTemplate(
-    input_variables=["context", "question"],
-    template=template,
-)
+        answer = call_deepseek(
+            messages=[
+                {"role": "system", "content": system_prompt},
+                {"role": "user", "content": user_prompt},
+            ]
+        )
 
-# 🔗 QA chain with custom prompt
-qa_chain = load_qa_chain(llm, chain_type="stuff", prompt=prompt)
+    except Exception as e:
+        print(f"Error while answering: {e}", flush=True)
+        answer = (
+            "Sorry, I ran into an error while answering. "
+            "Please try again in a moment."
+        )
 
-# 🔷 Question rephraser chain for follow-up questions → standalone
-CONDENSE_QUESTION_PROMPT = PromptTemplate.from_template(
-    """
-Given the following conversation and a follow-up question, rephrase the follow-up question to be a standalone question.
+    chat_history.append({"role": "user", "content": query})
+    chat_history.append({"role": "assistant", "content": answer})
 
-Chat History:
-{chat_history}
-Follow Up Input: {question}
-Standalone question:
-"""
-)
+    return "", chat_history
 
-question_generator = LLMChain(
-    llm=llm,
-    prompt=CONDENSE_QUESTION_PROMPT
-)
 
-# 🔷 Finally: build the ConversationalRetrievalChain manually
-chain = ConversationalRetrievalChain(
-    retriever=vectorstore.as_retriever(search_kwargs={'k': 6}),
-    question_generator=question_generator,
-    combine_docs_chain=qa_chain,
-    return_source_documents=True,
-    verbose=False
-)
+def clear_chat():
+    return [{"role": "assistant", "content": WELCOME_MESSAGE}]
+
+
+avatar_images = None
+guest_img = os.path.join(DOCS_DIR, "Guest.jpg")
+thierry_img = os.path.join(DOCS_DIR, "Thierry Picture.jpg")
+
+if os.path.exists(guest_img) and os.path.exists(thierry_img):
+    avatar_images = [guest_img, thierry_img]
 
-# 💬 Gradio UI
-chat_history = []
 
 with gr.Blocks() as demo:
+    gr.Markdown("# Thierry Decae Recruitment Chatbot")
+
     chatbot = gr.Chatbot(
-        [("", "Hello, I'm Thierry Decae's chatbot, you can ask me any recruitment related questions such as my experience, where I'm eligible to work, skills etc. You can chat with me directly in multiple languages.")],
-        avatar_images=["./multiple_docs/Guest.jpg", "./multiple_docs/Thierry Picture.jpg"]
+        value=[{"role": "assistant", "content": WELCOME_MESSAGE}],
+        type="messages",
+        avatar_images=avatar_images,
+        height=500,
     )
-    msg = gr.Textbox(placeholder="Type your question here...")
-    clear = gr.Button("Clear")
 
-    def user(query, chat_history):
-        chat_history_tuples = [(m[0], m[1]) for m in chat_history]
-        result = chain({"question": query, "chat_history": chat_history_tuples})
-        chat_history.append((query, result["answer"]))
-        return gr.update(value=""), chat_history
+    msg = gr.Textbox(
+        placeholder="Ask a recruitment-related question...",
+        label="Your question",
+    )
 
-    msg.submit(user, [msg, chatbot], [msg, chatbot], queue=False)
-    clear.click(lambda: None, None, chatbot, queue=False)
+    clear = gr.Button("Clear chat")
 
-demo.launch(debug=True)  # remove share=True if running in HF Spaces
+    msg.submit(
+        answer_question,
+        inputs=[msg, chatbot],
+        outputs=[msg, chatbot],
+    )
 
+    clear.click(
+        clear_chat,
+        inputs=None,
+        outputs=chatbot,
+    )
 
 
+demo.launch(
+    server_name="0.0.0.0",
+    server_port=int(os.getenv("PORT", 7860)),
+)