changes in datapreprocessing

app.py

@@ -3,9 +3,14 @@ import fitz
 import google.generativeai as genai
 from dotenv import load_dotenv
 import os
+from langchain.text_splitter import RecursiveCharacterTextSplitter
+from langchain.vectorstores import Chroma
+from langchain.embeddings import HuggingFaceEmbeddings
+from langchain.schema import Document
 
 load_dotenv()
-# Initialize the Google Gemini model
+
+# Initialize Google Gemini Model
 genai.configure(api_key=os.getenv("GOOGLE_API_KEY"))
 model = genai.GenerativeModel('gemini-1.5-flash')
 
@@ -17,6 +22,32 @@ def extract_text_from_pdf(file):
         text += page.get_text()
     return text
 
+def preprocess_and_chunk(resume_text):
+    """Preprocess and chunk the resume text for RAG-based retrieval."""
+    
+    # Wrap resume text in a Document object
+    docs = [Document(page_content=resume_text)]  # Convert the text into a Document object
+    
+    # Split text into chunks
+    text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000)
+    docs = text_splitter.split_documents(docs)
+    
+    # Create a vectorstore
+    vectorstore = Chroma(
+        collection_name="full_documents",
+        embedding_function=HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2",
+                                                 model_kwargs={'device': 'cpu'})
+    )
+    
+    vectorstore.add_documents(docs)
+    
+    return vectorstore
+
+def retrieve_relevant_text(query, vectorstore):
+    """Retrieve relevant text based on a query using the vector store."""
+    results = vectorstore.similarity_search(query)
+    return results
+
 def get_gemini_response(prompt):
     """Function to load Google Gemini model and provide queries as response."""
     response = model.generate_content([prompt])
@@ -24,34 +55,22 @@ def get_gemini_response(prompt):
 
 def analyze_resume(text):
     prompt = (
-    "You are an expert resume analyst with a focus on extracting precise, relevant information. Analyze the following resume text and provide the following detailed outputs:\n\n"
-    
-    "1. **Summary**:\n"
-    "   - Provide a concise and accurate summary of the resume.\n"
-    "   - Focus on the candidate's most important foundational skills, key technical competencies, and significant work experiences.\n"
-    "   - Highlight their primary areas of interest, notable projects, and any certifications directly relevant to their field.\n"
-    "   - Ensure that the summary directly aligns with the key details provided in the resume, avoiding any interpretation or unnecessary extrapolation. Maintain a neutral, objective tone.\n\n"
-    
-    "2. **Percentage Distribution of Fields/Domains**:\n"
-    "   - Break down the resume content into specific fields or domains, with a detailed percentage distribution.\n"
-    "   - Extract and list relevant keywords from each field/domain.\n"
-    "   - Ensure that the total distribution sums up to 100%.\n"
-    "   - Keywords should be limited to standard technical terms or industry-specific phrases, with no addition of non-essential details.\n"
-    "   - Ensure that the percentage distribution and keywords are strictly based on the content present in the resume text.\n\n"
-    
-    "### Example Format:\n\n"
-    "### Summary\n"
-    "This resume highlights a candidate with strong foundational skills in [Key Fields/Domains]. They have demonstrated technical proficiency in [Primary Technical Skills] and possess significant experience in [Type of Experience (e.g., internships, projects)]. The candidate has completed notable projects such as [Relevant Projects] and holds certifications in [Relevant Certifications]. Their primary areas of interest include [Primary Interests], demonstrating a strong alignment with their technical expertise.\n\n"
-    
-    "### Percentage Distribution of Fields/Domains\n"
-    "- **[Field/Domain 1]**: X%\n"
-    "  - Keywords: [Relevant Keywords]\n"
-    "- **[Field/Domain 2]**: Y%\n"
-    "  - Keywords: [Relevant Keywords]\n"
-    "- **[Field/Domain 3]**: Z%\n"
-    "  - Keywords: [Relevant Keywords]\n"
-)
-
+        "You are an expert resume analyst. Analyze the following resume text and provide the following:\n\n"
+        "1. A brief summary of the resume, including the candidate's main interests and the fields they are most passionate about.\n"
+        "2. A detailed percentage distribution of fields/domains present in the resume, with keywords extracted from the resume. Ensure the total sums up to 100%.\n"
+        "3. A overall explanation for each domain in brief according to the resume."
+        "Here is an example of how the output should look like:\n\n"
+        "### Summary\n"
+        "The resume indicates a strong background and interest in machine learning, data science, and software development. The candidate has worked on several projects involving machine learning algorithms, data preprocessing, and building software applications. They have demonstrated proficiency in Python, Java, and various machine learning frameworks. The candidate is passionate about solving complex problems using AI and has a keen interest in continuing to develop their skills in this area.\n\n"
+        "### Percentage Distribution of Fields/Domains\n"
+        "Note: only include standard technologies as keywords.\n"
+        "- *Machine Learning (ML)*: 40%\n"
+        "  - Keywords: Algorithms, Keras, PyTorch, Scikit-Learn, Predictive Models\n"
+        "- *Data Science (DS)*: 30%\n"
+        "  - Keywords: Data Analysis, Pandas, NumPy, Visualization, Statistical Methods\n"
+        "- *Software Development (SD)*: 30%\n"
+        "  - Keywords: Python, Java, Software Engineering, APIs, Git\n"
+    )
     analysis = get_gemini_response(prompt + text)
     return analysis
 
@@ -64,7 +83,7 @@ def extract_domains_from_analysis(analysis):
     )
     domain_response = get_gemini_response(domain_prompt)
     return domain_response.strip()
-
+    
 def generate_mcq_questions(analysis, selected_domain):
     num_questions = 20  # Default to 20 questions
     prompt_template = (
@@ -140,14 +159,17 @@ def generate_coding_questions(analysis, selected_domain):
 def generate_interview_questions(analysis, selected_domain):
     num_questions = 20  # Default to 20 questions
     prompt_template = (
-        f"Based on the candidate's resume and the identified skills, experience, and education, generate a set of {num_questions} interview questions "
+        f"Based on the candidate's resume {analysis} and the identified skills, experience, and education, generate a set of {num_questions} interview questions "
         f"that assess their fit for the position at our company. The questions should cover topics such as problem-solving abilities, leadership skills, "
         f"communication skills, cultural fit, etc. Additionally, include follow-up questions to probe deeper into the candidate's responses and evaluate their thought process. "
-        f"The questions should be specific to the field of {selected_domain}."
+        f"Note:The questions should be specific to the field of {selected_domain}."
+        f"Questions should be both knowledge base and industry application level also which include practical application of knowledge but only based on {selected_domain}."
+        f"You need to take reference from the resume analysis but the questions generated should be strictly based on {selected_domain}."
     )
     questions = get_gemini_response(prompt_template)
     return questions
 
+
 def split_questions_answers(quiz_response):
     """Function that splits the questions and answers from the quiz response."""
     if "Answers:" in quiz_response:
@@ -165,18 +187,80 @@ def main():
     if 'selected_domain' not in st.session_state:
         st.session_state.selected_domain = "DSA"
 
+    if 'analysis' not in st.session_state:
+        st.session_state.analysis = None
+
+    if 'uploaded_file' not in st.session_state:
+        st.session_state.uploaded_file = None
+
     uploaded_file = st.file_uploader("Upload your resume PDF", type="pdf")
-    if uploaded_file is not None:
+    if uploaded_file is not None and uploaded_file != st.session_state.uploaded_file:
+        st.session_state.uploaded_file = uploaded_file
         resume_text = extract_text_from_pdf(uploaded_file)
         
-        st.write("Analyzing Resume...")
-        analysis = analyze_resume(resume_text)
+        with st.spinner("Analyzing Resume..."):
+            st.session_state.analysis = analyze_resume(resume_text)
+        
+        st.write("Analysis Result:")
+        st.write(st.session_state.analysis)
         
+        # Extract domains from the analysis
+        domain_response = extract_domains_from_analysis(st.session_state.analysis)
+        domains = [domain.strip() for domain in domain_response.split(',')]
+        default_domains = ["DSA", "DBMS", "Programming Basics"] + domains
+
+        def update_selected_domain():
+            st.session_state.selected_domain = st.session_state.domain_select
+
+        st.selectbox(
+            "Select Domain for Questions:",
+            default_domains,
+            key="domain_select",
+            index=default_domains.index(st.session_state.selected_domain) if st.session_state.selected_domain in default_domains else 0,
+            on_change=update_selected_domain
+        )
+        
+        question_type = st.selectbox("Select Question Type:", ["Technical Round", "Interview Round"])
+        
+        if question_type == "Technical Round":
+            technical_subtype = st.selectbox("Select Technical Round Type:", ["MCQs", "Coding Challenges"])
+        
+        if st.button("Generate Questions"):
+            if question_type == "Technical Round":
+                if technical_subtype == "MCQs":
+                    quiz_response = generate_mcq_questions(st.session_state.analysis, st.session_state.selected_domain)
+                else:
+                    quiz_response = generate_coding_questions(st.session_state.analysis, st.session_state.selected_domain)
+            else:
+                quiz_response = generate_interview_questions(st.session_state.analysis, st.session_state.selected_domain)
+            
+            questions, answers = split_questions_answers(quiz_response)
+            
+            st.write("Generated Questions:")
+            st.write(questions)
+            
+            if st.button("Show Answers"):
+                st.write("Answers:")
+                st.write(answers)
+
+        # Process and store resume text in vectorstore
+        vectorstore = preprocess_and_chunk(resume_text)
+        
+        # Query for relevant text
+        query = st.text_input("Enter a query to search relevant text:", "")
+        if query:
+            with st.spinner("Retrieving relevant text..."):
+                relevant_texts = retrieve_relevant_text(query, vectorstore)
+                st.write("Relevant Texts:")
+                for text in relevant_texts:
+                    st.write(text.page_content)
+    
+    elif uploaded_file is not None and uploaded_file == st.session_state.uploaded_file:
         st.write("Analysis Result:")
-        st.write(analysis)
+        st.write(st.session_state.analysis)
         
         # Extract domains from the analysis
-        domain_response = extract_domains_from_analysis(analysis)
+        domain_response = extract_domains_from_analysis(st.session_state.analysis)
         domains = [domain.strip() for domain in domain_response.split(',')]
         default_domains = ["DSA", "DBMS", "Programming Basics"] + domains
 
@@ -199,11 +283,11 @@ def main():
         if st.button("Generate Questions"):
             if question_type == "Technical Round":
                 if technical_subtype == "MCQs":
-                    quiz_response = generate_mcq_questions(analysis, st.session_state.selected_domain)
+                    quiz_response = generate_mcq_questions(st.session_state.analysis, st.session_state.selected_domain)
                 else:
-                    quiz_response = generate_coding_questions(analysis, st.session_state.selected_domain)
+                    quiz_response = generate_coding_questions(st.session_state.analysis, st.session_state.selected_domain)
             else:
-                quiz_response = generate_interview_questions(analysis, st.session_state.selected_domain)
+                quiz_response = generate_interview_questions(st.session_state.analysis, st.session_state.selected_domain)
             
             questions, answers = split_questions_answers(quiz_response)
             
@@ -215,4 +299,4 @@ def main():
                 st.write(answers)
 
 if __name__ == "__main__":
-    main()
+    main()