Added metadata

app.py

@@ -90,7 +90,7 @@ def store_culture_docs(culture_files: List[tempfile._TemporaryFileWrapper]) -> s
     return f"Successfully stored {len(all_docs)} culture document chunks"
 
 def store_resumes(resume_files: List[tempfile._TemporaryFileWrapper]) -> str:
-    """Store resumes in the vector store"""
+    """Store resumes in the vector store with proper metadata"""
     text_splitter = RecursiveCharacterTextSplitter(
         chunk_size=1000,
         chunk_overlap=200
@@ -103,14 +103,23 @@ def store_resumes(resume_files: List[tempfile._TemporaryFileWrapper]) -> str:
         else:
             loader = UnstructuredFileLoader(file.name)
         docs = loader.load()
+        
+        # Extract filename without extension as resume ID
+        resume_id = os.path.splitext(os.path.basename(file.name))[0]
+        
+        # Add metadata to each chunk
         splits = text_splitter.split_documents(docs)
+        for split in splits:
+            split.metadata["resume_id"] = resume_id
+            split.metadata["source"] = "resume"
+        
         all_docs.extend(splits)
     
     resume_store.add_documents(all_docs)
-    return f"Successfully stored {len(all_docs)} resume chunks"
+    return f"Successfully stored {len(resume_files)} resumes"
 
 def analyze_candidates(job_description: str) -> str:
-    # Extract skills prompt template
+    # Extract skills first
     skills_prompt = PromptTemplate(
         input_variables=["job_description"],
         template="""
@@ -132,7 +141,7 @@ def analyze_candidates(job_description: str) -> str:
     # Query vector stores separately
     results = resume_store.similarity_search(
         job_description,
-        k=5
+        k=10  # Increase k to get more chunks
     )
     
     culture_docs = culture_store.similarity_search(
@@ -140,11 +149,25 @@ def analyze_candidates(job_description: str) -> str:
         k=3
     )
     
+    # Group resume chunks by resume_id
+    resume_groups = {}
+    for doc in results:
+        resume_id = doc.metadata.get("resume_id")
+        if resume_id not in resume_groups:
+            resume_groups[resume_id] = []
+        resume_groups[resume_id].append(doc.page_content)
+    
+    # Combine chunks for each resume
+    consolidated_resumes = {
+        resume_id: "\n".join(chunks)
+        for resume_id, chunks in resume_groups.items()
+    }
+    
     # Analysis prompt template
     analysis_prompt = PromptTemplate(
-        input_variables=["job_description", "skills", "culture_docs", "resumes"],
+        input_variables=["job_description", "skills", "culture_docs", "num_resumes", "resumes"],
         template="""
-        Analyze these candidates for the job position and culture fit.
+        Analyze the following {num_resumes} candidates for the job position and culture fit.
         
         Job Description:
         {job_description}
@@ -155,7 +178,7 @@ def analyze_candidates(job_description: str) -> str:
         Company Culture Context:
         {culture_docs}
         
-        Candidate Resumes:
+        Candidates to analyze:
         {resumes}
         
         For each candidate, provide:
@@ -163,6 +186,8 @@ def analyze_candidates(job_description: str) -> str:
         2. Culture fit assessment
         3. Recommendation (move forward/reject)
         4. Brief explanation
+        
+        Important: Only analyze the {num_resumes} distinct candidates provided above. Do not make up additional candidates.
         """
     )
     
@@ -171,13 +196,23 @@ def analyze_candidates(job_description: str) -> str:
         prompt=analysis_prompt
     )
     
+    # Format resumes for analysis
+    formatted_resumes = "\n\n".join([
+        f"Resume {i+1}:\n{content}"
+        for i, content in enumerate(consolidated_resumes.values())
+    ])
+    
     analysis = analysis_chain.run({
         "job_description": job_description,
         "skills": skills,
         "culture_docs": "\n".join([doc.page_content for doc in culture_docs]),
-        "resumes": "\n".join([doc.page_content for doc in results])
+        "num_resumes": len(consolidated_resumes),
+        "resumes": formatted_resumes
     })
     
+    print(f"Number of unique resumes found: {len(consolidated_resumes)}")
+    print(f"Resume IDs: {list(consolidated_resumes.keys())}")
+
     return analysis