Loan : Feature CV Interview questions

.gitignore

@@ -1,3 +1,7 @@
 .venv
 .env
-.store
+.store
+store
+*.pdf
+__pycache__
+*.pyc

main.py

@@ -9,6 +9,8 @@
 from modes.load_book_mode import LoadBookMode
 from modes.load_haiku_mode import LoadHaikuMode
 from modes.ask_mode import AskMode
+from modes.cv_mode import CVMode
+from modes.load_cv_mode import LoadCVMode
 
 load_dotenv()
 
@@ -34,5 +36,7 @@ def sigkill_handler(sig, frame):
     app.use("load-haiku", LoadHaikuMode)
     app.use("load-book", LoadBookMode)
     app.use("book", BookMode)
+    app.use("cv", CVMode)
+    app.use("load-cv", LoadCVMode)
 
     app.run()

modes/book_mode.py

@@ -7,7 +7,7 @@
 from langchain.chat_models import init_chat_model
 from langchain_core.messages import HumanMessage, AIMessage, BaseMessage
 from langchain_chroma import Chroma
-from langchain_openai import OpenAIEmbeddings
+from langchain_ollama import OllamaEmbeddings
 
 class BookMode(Mode):
 
@@ -16,7 +16,7 @@ class BookMode(Mode):
     def __init__(
         self, 
         console: Console,
-        model: str = "llama3.2:3b",
+        model: str = "llama3.2:1b",
         system: str = "default", 
         verbose: bool = False):
         super().__init__(console)
@@ -28,7 +28,7 @@ def __init__(
     @staticmethod
     def add_subparser(name: str, subparser: _SubParsersAction):
         chat_subparser = subparser.add_parser(name)
-        chat_subparser.add_argument("--model", type=str, default="llama3.2:3b")
+        chat_subparser.add_argument("--model", type=str, default="llama3.2:1b")
         chat_subparser.add_argument("--system", type=str, default="default")
         chat_subparser.add_argument("--verbose", "-v", action="store_true")
 
@@ -46,8 +46,8 @@ def run(self):
 
         # load VectorStore
         vector_store = Chroma(
-            embedding_function=OpenAIEmbeddings(),
-            persist_directory="./store"
+            embedding_function=OllamaEmbeddings(model="mxbai-embed-large:latest"),
+            persist_directory="./.store"
         )
 
         # Load model

modes/chat_mode.py

@@ -14,7 +14,7 @@ class ChatMode(Mode):
     def __init__(
         self, 
         console: Console,
-        model: str = "llama3.2:3b",
+        model: str = "llama3.2:1b",
         system: str = "default", 
         verbose: bool = False):
         super().__init__(console)
@@ -26,7 +26,7 @@ def __init__(
     @staticmethod
     def add_subparser(name: str, subparser: _SubParsersAction):
         chat_subparser = subparser.add_parser(name)
-        chat_subparser.add_argument("--model", type=str, default="llama3.2:3b")
+        chat_subparser.add_argument("--model", type=str, default="llama3.2:1b")
         chat_subparser.add_argument("--system", type=str, default="default")
         chat_subparser.add_argument("--verbose", "-v", action="store_true")
 

modes/cv_mode.py

@@ -0,0 +1,100 @@
+import os
+from mode import Mode
+from console import Console
+from argparse import _SubParsersAction
+from langchain_core.prompts import ChatPromptTemplate, SystemMessagePromptTemplate, MessagesPlaceholder
+from langchain_core.output_parsers import StrOutputParser
+from langchain.chat_models import init_chat_model
+from langchain_core.messages import HumanMessage, AIMessage, BaseMessage
+from langchain_chroma import Chroma
+from langchain_ollama import OllamaEmbeddings
+
+class CVMode(Mode):
+
+    history: list[BaseMessage] = []
+
+    def __init__(
+        self, 
+        console: Console,
+        model: str = "llama3.2:1b",
+        system: str = "default", 
+        verbose: bool = False):
+        super().__init__(console)
+
+        self.model = model
+        self.system = system
+        self.verbose = verbose
+
+    @staticmethod
+    def add_subparser(name: str, subparser: _SubParsersAction):
+        chat_subparser = subparser.add_parser(name)
+        chat_subparser.add_argument("--model", type=str, default="llama3.2:1b")
+        chat_subparser.add_argument("--system", type=str, default="default")
+        chat_subparser.add_argument("--verbose", "-v", action="store_true")
+
+    def run(self):
+        # System prompt
+        system_prompt = """
+            Tu es un recruteur expérimenté. Sur la base du CV fourni ci-dessous, rédige une série de questions d’entretien pertinentes pour évaluer le candidat.
+            Les questions doivent couvrir les aspects suivants :
+            - Questions techniques liées au domaine du candidat
+            - Questions comportementales (soft skills, travail en équipe, etc.)
+            - Questions de motivation ou de culture d’entreprise
+            Voici le CV de l'utilisateur :
+            {documents}
+        """
+
+        # Load VectorStore
+        vector_store = Chroma(
+            embedding_function=OllamaEmbeddings(model="mxbai-embed-large:latest"),
+            persist_directory="./.store"
+        )
+
+        # Requête pour orienter la recherche des documents pertinents
+        query = "Génère des questions d’entretien à partir du CV"
+        documents = vector_store.similarity_search(query, k=4)
+
+        # Affichage des documents récupérés (optionnel)
+        if self.verbose:
+            self.console.info("Documents chargés :")
+            for document in documents:
+                self.console.info(document.page_content)
+
+        # Chargement du modèle
+        if self.verbose:
+            self.console.info(f"Chargement du modèle {self.model}...")
+
+        model = init_chat_model(
+            self.model,
+            model_provider="ollama",
+            temperature=0.7,
+        )
+
+        # Construction du prompt
+        prompt = ChatPromptTemplate.from_messages([
+            SystemMessagePromptTemplate.from_template(system_prompt),
+            MessagesPlaceholder(variable_name="messages"),
+        ])
+
+        # Construction de la chaîne
+        chain = prompt | model | StrOutputParser()
+
+        # Message utilisateur simulé pour lancer la génération
+        user_message = HumanMessage(content="Génère des questions d’entretien adaptées à mon CV.")
+        self.history.append(user_message)
+
+        # Exécution de la chaîne
+        self.console.bot_start()
+        stream = chain.stream({
+            "messages": self.history,
+            "documents": documents
+        })
+
+        # Affichage du flux de réponse
+        bot_message = ""
+        for chunk in stream:
+            bot_message += chunk
+            self.console.bot_chunk(chunk)
+        self.console.bot_end()
+
+        self.history.append(AIMessage(bot_message))

modes/load_book_mode.py

@@ -3,7 +3,7 @@
 from mode import Mode
 from langchain_community.document_loaders import PyPDFLoader
 from langchain_experimental.text_splitter import SemanticChunker
-from langchain_openai import OpenAIEmbeddings
+from langchain_ollama import OllamaEmbeddings
 from langchain_core.documents import Document
 from langchain_chroma import Chroma
 
@@ -30,18 +30,19 @@ def run(self):
 
         # Create vector store
         vector_store = Chroma(
-            embedding_function=OpenAIEmbeddings(),
-            persist_directory="./store"
+            embedding_function=OllamaEmbeddings(model="mxbai-embed-large:latest"),
+            persist_directory="./.store"
         )
 
         # Loading
         text_splitter = SemanticChunker(
-            embeddings=OpenAIEmbeddings(),
+            embeddings=OllamaEmbeddings(model="mxbai-embed-large:latest"),
         )
 
         loader = PyPDFLoader(self.book)
         for page in loader.lazy_load():
             chunks = text_splitter.split_documents([page])
+            print(f"Loaded {len(chunks)} chunks from page {page.metadata['page']}")
             vector_store.add_documents(chunks)
 
 

modes/load_cv_mode.py

@@ -0,0 +1,47 @@
+from argparse import _SubParsersAction
+from console import Console
+from mode import Mode
+from langchain_community.document_loaders import PyPDFLoader
+from langchain_experimental.text_splitter import SemanticChunker
+from langchain_ollama import OllamaEmbeddings
+from langchain_core.documents import Document
+from langchain_chroma import Chroma
+
+
+class LoadCVMode(Mode):
+    def __init__(
+        self, 
+        console: Console, 
+        cv: str, 
+        verbose: bool = False):
+        super().__init__(console)
+
+        self.cv = cv
+        self.verbose = verbose
+
+    @staticmethod
+    def add_subparser(name: str, subparser: _SubParsersAction):
+        load_cv_subparser = subparser.add_parser(name)
+        load_cv_subparser.add_argument("cv", type=str, help="The cv to load")
+        load_cv_subparser.add_argument("--verbose", "-v", action="store_true", help="Verbose mode")
+
+    def run(self):
+        self.console.info(f"Loading cv {self.cv}...")
+
+        # Create vector store
+        vector_store = Chroma(
+            embedding_function=OllamaEmbeddings(model="mxbai-embed-large:latest"),
+            persist_directory="./.store"
+        )
+
+        # Loading
+        text_splitter = SemanticChunker(
+            embeddings=OllamaEmbeddings(model="mxbai-embed-large:latest"),
+        )
+
+        loader = PyPDFLoader(self.cv)
+        for page in loader.lazy_load():
+            chunks = text_splitter.split_documents([page])
+            vector_store.add_documents(chunks)
+
+