Update app.py

app.py

@@ -5,24 +5,22 @@ import os
 from huggingface_hub import login
 import spacy
 
-# ✅ Authenticate using HF token stored in secret
+# Authenticate with Hugging Face token
 login(token=os.environ.get("HUGGINGFACEHUB_API_TOKEN"))
 
-# ✅ Load summarization model (lightweight and CPU-friendly)
+# Load models
 summarizer = pipeline("text2text-generation", model="declare-lab/flan-alpaca-base")
-
-# ✅ Load SpaCy English model for NLP keyword extraction
 nlp = spacy.load("en_core_web_sm")
 
-# ✅ Extract keywords (nouns & proper nouns, excluding stopwords)
+# Extract contextually relevant keywords using spaCy
 def extract_relevant_keywords(text):
     doc = nlp(text.lower())
     return set(
-        token.text for token in doc
+        token.lemma_ for token in doc
         if token.pos_ in {"NOUN", "PROPN"} and not token.is_stop and len(token.text) > 2
     )
 
-# ✅ Match resume keywords with job description
+# Compare resume and JD for keyword matches
 def compare_keywords(resume_text, job_desc):
     resume_words = extract_relevant_keywords(resume_text)
     job_words = extract_relevant_keywords(job_desc)
@@ -30,46 +28,45 @@ def compare_keywords(resume_text, job_desc):
     missing = job_words - resume_words
     return matched, missing
 
-# ✅ Highlight matched keywords in resume text
+# Highlight matched words in resume text
 def highlight_keywords(resume_text, matched):
     highlighted = resume_text
     for word in sorted(matched, key=len, reverse=True):
         highlighted = re.sub(rf"\b({re.escape(word)})\b", r"**\1**", highlighted, flags=re.IGNORECASE)
     return highlighted
 
-# ✅ Use LLM to extract contextually missing skills/tools
+# LLM-based keyword extraction from JD
 def extract_missing_keywords_with_llm(job_desc, resume_text):
     prompt = f"""
 Given the following job description and resume, list the important skills, tools, and concepts from the job description that are missing or weakly represented in the resume.
+Only list the missing keywords as bullet points.
 
 Job Description:
 {job_desc}
 
 Resume:
 {resume_text}
-
-Only list the missing keywords as bullet points.
 """
-    result = summarizer(prompt, max_new_tokens=300, do_sample=True)[0]
-    return result.get('generated_text', str(result)).strip()
+    result = summarizer(prompt, max_new_tokens=300, do_sample=False)[0]
+    raw_text = result.get('generated_text', result.get('summary_text', str(result))).strip()
+
+    # Clean and deduplicate
+    lines = re.findall(r"-\s*(.+)", raw_text)
+    cleaned = list({kw.strip().lower() for kw in lines if len(kw.strip()) > 2})
+    return ', '.join(sorted(cleaned)) or "None"
 
-# ✅ Build LLM prompt to extract resume sections + insights
+# Prompt builder for structured LLM resume analysis
 def build_dynamic_prompt(job_desc, resume_text, analyze_with_jd):
     prompt = f"""
-You are a professional resume analyst.
-
-Classify the content of the resume below into logical sections such as:
-- Education
-- Work Experience
+Act as a resume evaluator. Break the following resume into meaningful sections such as:
 - Technical Skills
 - Soft Skills
+- Education
+- Experience
 - Certifications
-- Projects
-- Achievements
-
-Use markdown format with headers (###) for each section.
+- Projects (if present)
 
-If a job description is also provided, add relevant suggestions under each section to improve alignment with the job.
+Then, if a job description is provided, highlight what improvements are needed in each section to better align with the job role.
 
 Job Description:
 {job_desc if analyze_with_jd else '[None provided]'}
@@ -77,29 +74,27 @@ Job Description:
 Resume:
 {resume_text}
 
-Return your output in structured markdown format.
+Output your response in markdown format with section headings.
 """
     return prompt
 
-# ✅ Generate structured LLM output + keyword analysis
+# Core analysis function
 def analyze_resume(job_desc, resume_text, analyze_with_jd):
     if not resume_text.strip():
         return "⚠️ Please paste your resume text."
 
-    prompt = build_dynamic_prompt(job_desc, resume_text, analyze_with_jd)
+    user_prompt = build_dynamic_prompt(job_desc, resume_text, analyze_with_jd)
 
     try:
-        llm_result = summarizer(prompt, max_new_tokens=512, do_sample=True)[0]
-        structured_response = llm_result.get('generated_text', str(llm_result)).strip()
+        result = summarizer(user_prompt, max_new_tokens=512, do_sample=False)[0]
+        response_text = result.get('generated_text', result.get('summary_text', str(result))).strip()
 
-        # Perform keyword comparison and highlighting
         if analyze_with_jd and job_desc:
             matched, missing = compare_keywords(resume_text, job_desc)
             highlighted_resume = highlight_keywords(resume_text, matched)
             llm_missing_keywords = extract_missing_keywords_with_llm(job_desc, resume_text)
 
-            return f"""
-### 🔍 Resume with Highlighted Matches
+            return f"""### 🔍 Resume with Highlighted Matches
 
 {highlighted_resume}
 
@@ -116,14 +111,12 @@ def analyze_resume(job_desc, resume_text, analyze_with_jd):
 
 ---
 
-{structured_response}
-"""
-        return structured_response
-
+{response_text}"""
+        return response_text
     except Exception as e:
         return f"❌ Error: {str(e)}"
 
-# ✅ Gradio UI
+# Gradio interface
 def create_ui():
     with gr.Blocks() as demo:
         with gr.Row():
@@ -141,3 +134,4 @@ def create_ui():
 
 if __name__ == '__main__':
     create_ui().launch()
+