Update app.py

app.py

@@ -2,7 +2,7 @@ import os
 import sys
 import requests
 
-# SQLite workaround for Chroma on Hugging Face Spaces
+# SQLite workaround (needed for Chroma on HF Spaces)
 try:
     __import__("pysqlite3")
     sys.modules["sqlite3"] = sys.modules.pop("pysqlite3")
@@ -17,9 +17,11 @@ from langchain_chroma import Chroma
 from langchain_huggingface import HuggingFaceEmbeddings
 
 
+# ========================
+# CONFIG
+# ========================
 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"
@@ -31,9 +33,12 @@ WELCOME_MESSAGE = (
 )
 
 
-def call_deepseek(messages, temperature=0.4, max_tokens=700):
+# ========================
+# DEEPSEEK CALL
+# ========================
+def call_deepseek(messages):
     if not DEEPSEEK_API_KEY:
-        return "DEEPSEEK_API_KEY is not set in the Hugging Face Space secrets."
+        return "Missing DEEPSEEK_API_KEY."
 
     headers = {
         "Authorization": f"Bearer {DEEPSEEK_API_KEY}",
@@ -43,211 +48,166 @@ def call_deepseek(messages, temperature=0.4, max_tokens=700):
     payload = {
         "model": "deepseek-chat",
         "messages": messages,
-        "temperature": temperature,
-        "max_tokens": max_tokens,
+        "temperature": 0.4,
+        "max_tokens": 700,
     }
 
-    response = requests.post(
-        DEEPSEEK_API_URL,
-        headers=headers,
-        json=payload,
-        timeout=60,
-    )
+    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()
+    return response.json()["choices"][0]["message"]["content"].strip()
 
 
+# ========================
+# LOAD DOCS
+# ========================
 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()
+    for f in os.listdir(DOCS_DIR):
+        path = os.path.join(DOCS_DIR, f)
 
         try:
-            if lower.endswith(".pdf"):
-                loader = PyPDFLoader(path)
-                docs.extend(loader.load())
-
-            elif lower.endswith(".docx"):
-                loader = Docx2txtLoader(path)
-                docs.extend(loader.load())
-
-            elif lower.endswith(".txt"):
-                loader = TextLoader(path, encoding="utf-8")
-                docs.extend(loader.load())
-
+            if f.endswith(".pdf"):
+                docs.extend(PyPDFLoader(path).load())
+            elif f.endswith(".docx"):
+                docs.extend(Docx2txtLoader(path).load())
+            elif f.endswith(".txt"):
+                docs.extend(TextLoader(path, encoding="utf-8").load())
         except Exception as e:
-            print(f"Could not load {filename}: {e}", flush=True)
+            print(f"Error loading {f}: {e}", flush=True)
 
     if not docs:
-        raise ValueError(f"No documents found in {DOCS_DIR}")
-
-    splitter = CharacterTextSplitter(
-        chunk_size=1000,
-        chunk_overlap=100,
-    )
+        raise ValueError("No documents found")
 
+    splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=100)
     return splitter.split_documents(docs)
 
 
+# ========================
+# VECTORSTORE
+# ========================
 def build_vectorstore():
-    print("Loading embedding model...", flush=True)
+    print("Loading embeddings...", flush=True)
 
-    embedding_function = HuggingFaceEmbeddings(
+    embedding = HuggingFaceEmbeddings(
         model_name="sentence-transformers/all-MiniLM-L6-v2"
     )
 
-    print("Loading documents...", flush=True)
     docs = load_documents()
+    print(f"Loaded {len(docs)} chunks", flush=True)
 
-    print(f"Loaded {len(docs)} document chunks.", flush=True)
-
-    vectorstore = Chroma.from_documents(
-        documents=docs,
-        embedding=embedding_function,
+    return Chroma.from_documents(
+        docs,
+        embedding,
         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):
+# ========================
+# HISTORY FORMAT
+# ========================
+def format_history(history):
     if not history:
         return ""
 
     lines = []
-
-    for item in history[-8:]:
-        if not item or len(item) != 2:
-            continue
-
-        user_msg, assistant_msg = item
-
-        if user_msg:
-            lines.append(f"user: {user_msg}")
-
-        if assistant_msg:
-            lines.append(f"assistant: {assistant_msg}")
+    for msg in history[-8:]:
+        role = msg.get("role")
+        content = msg.get("content")
+        if role and content:
+            lines.append(f"{role}: {content}")
 
     return "\n".join(lines)
 
 
-def answer_question(query, chat_history):
-    if chat_history is None:
-        chat_history = [[None, WELCOME_MESSAGE]]
-
-    if not query or not query.strip():
-        return "", chat_history
+# ========================
+# MAIN QA FUNCTION
+# ========================
+def answer_question(query, history):
+    if history is None:
+        history = [{"role": "assistant", "content": WELCOME_MESSAGE}]
 
-    query = query.strip()
+    if not query.strip():
+        return "", history
 
     try:
-        retrieved_docs = retriever.invoke(query)
+        docs = retriever.invoke(query)
+        context = "\n\n".join(d.page_content for d in docs if d.page_content)
 
-        context = "\n\n".join(
-            doc.page_content for doc in retrieved_docs if doc.page_content
-        )
-
-        history_text = format_chat_history(chat_history)
+        history_text = format_history(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.
+Answer questions about Thierry's experience, skills, and career.
+Use only provided context.
+If unsure, say "I'm not sure about that."
+Always answer as Thierry ("I", "my").
 """
 
         user_prompt = f"""
-Conversation history:
+Conversation:
 {history_text}
 
-Context from Thierry's documents:
+Context:
 {context}
 
-User question:
+Question:
 {query}
 
 Answer:
 """
 
-        answer = call_deepseek(
-            messages=[
-                {"role": "system", "content": system_prompt},
-                {"role": "user", "content": user_prompt},
-            ]
-        )
+        answer = call_deepseek([
+            {"role": "system", "content": system_prompt},
+            {"role": "user", "content": user_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."
-        )
+        print(e, flush=True)
+        answer = "Error while answering."
 
-    chat_history.append([query, answer])
+    history.append({"role": "user", "content": query})
+    history.append({"role": "assistant", "content": answer})
 
-    return "", chat_history
+    return "", history
 
 
 def clear_chat():
-    return [[None, WELCOME_MESSAGE]]
+    return [{"role": "assistant", "content": WELCOME_MESSAGE}]
 
 
-avatar_images = None
-
+# ========================
+# UI
+# ========================
 guest_img = os.path.join(DOCS_DIR, "Guest.jpg")
 thierry_img = os.path.join(DOCS_DIR, "Thierry Picture.jpg")
 
+avatars = None
 if os.path.exists(guest_img) and os.path.exists(thierry_img):
-    avatar_images = [guest_img, thierry_img]
+    avatars = [guest_img, thierry_img]
 
 
 with gr.Blocks() as demo:
     gr.Markdown("# Thierry Decae Recruitment Chatbot")
 
     chatbot = gr.Chatbot(
-        value=[[None, WELCOME_MESSAGE]],
-        avatar_images=avatar_images,
+        value=[{"role": "assistant", "content": WELCOME_MESSAGE}],
+        avatar_images=avatars,
         height=500,
     )
 
-    msg = gr.Textbox(
-        placeholder="Ask a recruitment-related question...",
-        label="Your question",
-    )
+    msg = gr.Textbox(placeholder="Ask a question...")
 
-    clear = gr.Button("Clear chat")
+    clear = gr.Button("Clear")
 
-    msg.submit(
-        answer_question,
-        inputs=[msg, chatbot],
-        outputs=[msg, chatbot],
-    )
-
-    clear.click(
-        clear_chat,
-        inputs=None,
-        outputs=chatbot,
-    )
+    msg.submit(answer_question, [msg, chatbot], [msg, chatbot])
+    clear.click(clear_chat, None, chatbot)
 
 
 demo.launch(