Update app.py

app.py

@@ -6,7 +6,6 @@ from tavily import TavilyClient
 from pydantic import BaseModel
 from ollama import chat
 from dotenv import load_dotenv
-from groq import Groq
 import instructor
 import logging
 from together import Together
@@ -16,12 +15,26 @@ import json
 logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
 logger = logging.getLogger(__name__)
 
-GROQ_API_KEY = "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"
+
 ELEVEN_LABS_API_KEY = "sk_cc3fea7dcfd81744dcc51673fcd011e7315d4732bab408a7"
 TAVILY_API_KEY = "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"
 
 load_dotenv()
 
+def call_llm(prompt):
+    client = Together(api_key = "aa77adf5b5adaefe8fb3e4a5a1e9bb4937ba9d5d362e03de2521631ab9dab07f")
+    response = client.chat.completions.create(
+            model="meta-llama/Llama-3.3-70B-Instruct-Turbo",
+            messages=[
+            {
+                "role": "user",
+                "content": prompt
+            }
+            ]
+        )
+    response = response.choices[0].message.content
+
+    return response
 
 def fetch_from_web(query):
     tavily_client = TavilyClient(api_key=TAVILY_API_KEY)
@@ -42,7 +55,7 @@ class Sentiment(BaseModel):
     sentiment: Literal['positive', 'negative', 'neutral']
 
 
-def analyze_sentiment(article, model_provider = "Groq"):
+def analyze_sentiment(article):
     sentiment_prompt = f"""
                 Analyze the following news article about a company:
 
@@ -65,78 +78,30 @@ def analyze_sentiment(article, model_provider = "Groq"):
                 """
 
     try:
-        if model_provider == "Ollama":
-            response = chat(
-                messages=[
-                    {
-                        'role': 'user',
-                        'content': sentiment_prompt
-                    }
-                ],
-                model='llama3.2:3b',
-                format=Sentiment.model_json_schema(),
-            )
+        client = Together(api_key = "aa77adf5b5adaefe8fb3e4a5a1e9bb4937ba9d5d362e03de2521631ab9dab07f")
 
-            sentiment_output = Sentiment.model_validate_json("")
-            
-            final_dict = {
-                "title": article["title"],
-                "summary": sentiment_output.summary,
-                "reasoning": sentiment_output.reasoning,
-                "topics": sentiment_output.topics,
-                "sentiment": sentiment_output.sentiment
-            }
-        elif model_provider == "GROQ":
-            llm = Groq(api_key=GROQ_API_KEY)
-            llm = instructor.from_groq(llm, mode=instructor.Mode.TOOLS)
-
-            resp = llm.chat.completions.create(
-                model="llama-3.3-70b-versatile",
-                messages=[
-                    {
-                        "role": "user",
-                        "content": sentiment_prompt,
-                    }
-                ],
-                response_model=Sentiment,
-            )
-        
-            sentiment_output = resp.model_dump()
-            
-            final_dict = {
-                "title": article["title"],
-                "summary": sentiment_output.get("summary"),
-                "reasoning": sentiment_output.get("reasoning"),
-                "topics": sentiment_output.get("topics"),
-                "sentiment": sentiment_output.get("sentiment")
-            }
-
-        elif model_provider == "Together":
-
-            client = Together(api_key = "aa77adf5b5adaefe8fb3e4a5a1e9bb4937ba9d5d362e03de2521631ab9dab07f")
-
-            extract = client.chat.completions.create(
-            messages=[
+        extract = client.chat.completions.create(
+        messages=[
                     {
                         "role": "user",
                         "content": sentiment_prompt,
                     },
                 ],
-                model="meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo",
+                model="meta-llama/Llama-3.3-70B-Instruct-Turbo",
                 response_format={
                     "type": "json_object",
                     "schema": Sentiment.model_json_schema(),
                 },
             )
 
-            output = json.loads(extract.choices[0].message.content)
+        output = json.loads(extract.choices[0].message.content)
            
-            final_dict = {
+        final_dict = {
                 "title": article["title"],
-                "summary": output .get("summary"),
-                "reasoning": output .get("reasoning"),
-                "topics": output .get("topics"),
-                "sentiment": output .get("sentiment")
+                "summary": output.get("summary"),
+                "reasoning": output.get("reasoning"),
+                "topics": output.get("topics"),
+                "sentiment": output.get("sentiment")
             }
 
         return final_dict
@@ -227,146 +192,96 @@ def get_summaries_by_sentiment(articles):
     return pos_sum, neg_sum, neutral_sum
 
 
-def comparative_analysis(pos_sum, neg_sum, neutral_sum, model_provider = "Groq"):
+def comparative_analysis(pos_sum, neg_sum, neutral_sum):
     prompt = f"""
-Perform a detailed comparative analysis of the sentiment across three categories of articles (Positive, Negative, and Neutral) about a specific company. Address the following aspects:
-
-1. **Sentiment Breakdown**: Identify how each category (positive, negative, and neutral) portrays the company. Highlight the language, tone, and emotional cues that shape the sentiment.
+    Perform a detailed comparative analysis of the sentiment across three categories of articles (Positive, Negative, and Neutral) about a specific company. Address the following aspects:
 
-2. **Key Themes and Topics**: Compare the primary themes and narratives within each sentiment group. What aspects of the company's operations, performance, or reputation does each category focus on?
+    1. **Sentiment Breakdown**: Identify how each category (positive, negative, and neutral) portrays the company. Highlight the language, tone, and emotional cues that shape the sentiment.
 
-3. **Perceived Company Image**: Analyze how each sentiment type influences public perception of the company. What impression is created by positive vs. negative vs. neutral coverage?
+    2. **Key Themes and Topics**: Compare the primary themes and narratives within each sentiment group. What aspects of the company's operations, performance, or reputation does each category focus on?
 
-4. **Bias and Framing**: Evaluate whether any of the articles reflect explicit biases or specific agendas regarding the company. Are there patterns in how the company is framed across different sentiments?
+    3. **Perceived Company Image**: Analyze how each sentiment type influences public perception of the company. What impression is created by positive vs. negative vs. neutral coverage?
 
-5. **Market or Stakeholder Impact**: Discuss potential effects on stakeholders (e.g., investors, customers, regulators) based on the sentiment of each article type.
+    4. **Bias and Framing**: Evaluate whether any of the articles reflect explicit biases or specific agendas regarding the company. Are there patterns in how the company is framed across different sentiments?
 
-6. **Comparative Insights**: Provide a concise summary of the major differences and commonalities between the three sentiment groups. What overall narrative emerges about the company?
+    5. **Market or Stakeholder Impact**: Discuss potential effects on stakeholders (e.g., investors, customers, regulators) based on the sentiment of each article type.
 
-### Positive Articles:
-{pos_sum}
+    6. **Comparative Insights**: Provide a concise summary of the major differences and commonalities between the three sentiment groups. What overall narrative emerges about the company?
 
-### Negative Articles:
-{neg_sum}
+    ### Positive Articles:
+    {pos_sum}
 
-### Neutral Articles:
-{neutral_sum}
-"""
+    ### Negative Articles:
+    {neg_sum}
 
-    if model_provider == "Ollama":
-        response = chat(
-            messages=[
-                {
-                    'role': 'user',
-                    'content': prompt
-                }
-            ],
-            model='llama3.2:3b'
-        )
-        response = response.message.content
-    
-    else:
-        llm = Groq(api_key=GROQ_API_KEY)
-        
-        chat_completion = llm.chat.completions.create(
-            messages=[
-                {
-                    "role": "user",
-                    "content": prompt[:5000],
-                }
-            ],
-            model="llama-3.3-70b-versatile",
-        )
-        response = chat_completion.choices[0].message.content
-    
-    return response
+    ### Neutral Articles:
+    {neutral_sum}
+    """
 
+    output = call_llm(prompt)
+    return output
 
-def generate_final_report(pos_sum, neg_sum, neutral_sum, comparative_sentiment, model_provider = "Groq"):
+def generate_final_report(pos_sum, neg_sum, neutral_sum, comparative_sentiment):
     final_report_prompt = f"""
-    Corporate News Sentiment Analysis Report:
-
-### 1. Executive Summary
-- Overview of sentiment distribution: {comparative_sentiment["Sentiment Distribution"]['Positive']} positive, {comparative_sentiment["Sentiment Distribution"]['Negative']} negative, {comparative_sentiment["Sentiment Distribution"]['Neutral']} neutral.
-- Highlight the dominant narrative shaping the company's perception.
-- Summarize key drivers behind positive and negative sentiments.
-
-### 2. Media Coverage Analysis
-- Identify major news sources covering the company.
-- Highlight patterns in coverage across platforms (e.g., frequency, timing).
-- Identify whether media sentiment shifts over time.
-
-### 3. Sentiment Breakdown
-- **Positive Sentiment:**
-    * Titles and sources: {pos_sum}
-    * Key themes, notable quotes, and focal areas (e.g., product, leadership).
-- **Negative Sentiment:**
-    * Titles and sources: {neg_sum}
-    * Key themes, notable quotes, and areas of concern.
-- **Neutral Sentiment:**
-    * Titles and sources: {neutral_sum}
-    * Key themes and neutral narratives.
-
-### 4. Narrative Analysis
-- Identify primary storylines about the company.
-- Analyze how the company is positioned (positive, neutral, negative).
-- Detect shifts or emerging narratives over time.
-
-### 5. Key Drivers of Sentiment
-- Identify specific events, announcements, or actions driving media sentiment.
-- Evaluate sentiment linked to industry trends vs. company-specific factors.
-- Highlight company strengths and weaknesses based on media portrayal.
-
-### 6. Competitive Context
-- Identify competitor comparisons.
-- Analyze how media sentiment about the company compares to industry standards.
-- Highlight competitive advantages or concerns raised by the media.
-
-### 7. Stakeholder Perspective
-- Identify how key stakeholders (e.g., investors, customers, regulators) are represented.
-- Analyze stakeholder concerns and reputation risks/opportunities.
-
-### 8. Recommendations
-- Suggest strategies to mitigate negative sentiment.
-- Recommend approaches to amplify positive narratives.
-- Provide messaging suggestions for future announcements.
-
-### 9. Appendix
-- Full article details (title, publication, date, author, URL).
-- Sentiment scoring methodology.
-- Media monitoring metrics (reach, engagement, etc.).
-"""
-
-    if model_provider == "Ollama":
-        final_report = chat(
-            messages=[
-                {
-                    'role': 'user',
-                    'content': final_report_prompt
-                }
-            ],
-            model='llama3.2:3b'
-        )
-        response = final_report.message.content
-    
-    else:
-        llm = Groq(api_key=GROQ_API_KEY)
-        
-        chat_completion = llm.chat.completions.create(
-            messages=[
-                {
-                    "role": "user",
-                    "content": final_report_prompt[:5000],
-                }
-            ],
-            model="llama-3.3-70b-versatile",
-        )
-        response = chat_completion.choices[0].message.content
+        Corporate News Sentiment Analysis Report:
+
+    ### 1. Executive Summary
+    - Overview of sentiment distribution: {comparative_sentiment["Sentiment Distribution"]['Positive']} positive, {comparative_sentiment["Sentiment Distribution"]['Negative']} negative, {comparative_sentiment["Sentiment Distribution"]['Neutral']} neutral.
+    - Highlight the dominant narrative shaping the company's perception.
+    - Summarize key drivers behind positive and negative sentiments.
+
+    ### 2. Media Coverage Analysis
+    - Identify major news sources covering the company.
+    - Highlight patterns in coverage across platforms (e.g., frequency, timing).
+    - Identify whether media sentiment shifts over time.
+
+    ### 3. Sentiment Breakdown
+    - **Positive Sentiment:**
+        * Titles and sources: {pos_sum}
+        * Key themes, notable quotes, and focal areas (e.g., product, leadership).
+    - **Negative Sentiment:**
+        * Titles and sources: {neg_sum}
+        * Key themes, notable quotes, and areas of concern.
+    - **Neutral Sentiment:**
+        * Titles and sources: {neutral_sum}
+        * Key themes and neutral narratives.
+
+    ### 4. Narrative Analysis
+    - Identify primary storylines about the company.
+    - Analyze how the company is positioned (positive, neutral, negative).
+    - Detect shifts or emerging narratives over time.
+
+    ### 5. Key Drivers of Sentiment
+    - Identify specific events, announcements, or actions driving media sentiment.
+    - Evaluate sentiment linked to industry trends vs. company-specific factors.
+    - Highlight company strengths and weaknesses based on media portrayal.
+
+    ### 6. Competitive Context
+    - Identify competitor comparisons.
+    - Analyze how media sentiment about the company compares to industry standards.
+    - Highlight competitive advantages or concerns raised by the media.
+
+    ### 7. Stakeholder Perspective
+    - Identify how key stakeholders (e.g., investors, customers, regulators) are represented.
+    - Analyze stakeholder concerns and reputation risks/opportunities.
+
+    ### 8. Recommendations
+    - Suggest strategies to mitigate negative sentiment.
+    - Recommend approaches to amplify positive narratives.
+    - Provide messaging suggestions for future announcements.
+
+    ### 9. Appendix
+    - Full article details (title, publication, date, author, URL).
+    - Sentiment scoring methodology.
+    - Media monitoring metrics (reach, engagement, etc.).
+    """
+
+    response = call_llm(final_report_prompt)
     
     return response
     
 
-def translate(report, model_provider = "Groq"):
+def translate(report):
     translation_prompt = f"""
     Translate the following corporate sentiment analysis report into Hindi:
 
@@ -374,33 +289,8 @@ def translate(report, model_provider = "Groq"):
 
     Ensure the translation maintains professional tone and structure while accurately conveying key insights and details.
     """
-    if model_provider == "Ollama":
-        translation = chat(
-            messages=[
-                {
-                    'role': 'user',
-                    'content': translation_prompt
-                }
-            ],
-            model='llama3.2:3b'
-        )
-        response = translation.message.content
-    
-    else:
-        translation_llm = Groq(api_key=GROQ_API_KEY)
-        
-        chat_completion = translation_llm.chat.completions.create(
-            messages=[
-                {
-                    "role": "user",
-                    "content": translation_prompt[:5000],
-                }
-            ],
-            model="llama-3.3-70b-versatile",
-        )
-        response = chat_completion.choices[0].message.content
-        
-    return response 
+    translation = call_llm(translation_prompt)
+    return translation
 
 
 def text_to_speech(text):
@@ -447,9 +337,10 @@ if st.button("Fetch Sentiment Data"):
             st.error("No sources found.")
         else:
             sentiment_output = [
-                analyze_sentiment(article , model_provider = "Together")
+                analyze_sentiment(article)
                 for article in web_results["sources"][:5]
             ]
+            sentiment_output = [s for s in sentiment_output if s is not None]
             logger.info(f"Generating comparative sentiment")
             comparative_sentiment = generate_comparative_sentiment(sentiment_output)
 
@@ -463,15 +354,13 @@ if st.button("Fetch Sentiment Data"):
                 positive_summary,
                 negative_summary,
                 neutral_summary,
-                comparative_sentiment,
-                model_provider = "Groq"
-                
+                comparative_sentiment   
             )
 
             logger.info(f"Translating Report")
-            hindi_translation = translate(final_report , model_provider = "Groq")
+            hindi_translation = translate(final_report)
 
-            logger.info(f"Generating Speech from Text")
+            #logger.info(f"Generating Speech from Text")
             #audio_data = text_to_speech(hindi_translation)
 
             output_dict = {