Use system instructions to steer the behavior of a model

lib/ai_writers/blog_from_google_serp.py

@@ -15,26 +15,9 @@ from ..gpt_providers.text_generation.main_text_generation import llm_text_gen
 
 def write_blog_google_serp(search_keyword, search_results):
     """Combine the given online research and GPT blog content"""
-    try:
-        config_path = Path(os.environ["ALWRITY_CONFIG"])
-        with open(config_path, 'r', encoding='utf-8') as file:
-            config = json.load(file)
-    except Exception as err:
-        logger.error(f"Error: Failed to read values from config: {err}")
-        exit(1)
-
-    blog_characteristics = config['Blog Content Characteristics']
-    
     prompt = f"""
         As expert Creative Content writer,
-        I want you to write {blog_characteristics['Blog Type']} blog post,
+        I want you to write blog post, that explores {search_keyword} and also include 5 FAQs.
-        that explores {search_keyword} and also include 5 FAQs.
-        
-        Below are the guidelines to follow:
-        1). You must respond in {blog_characteristics['Blog Language']} language.
-        2). Tone and Brand Alignment: Adjust your tone, voice, personality for {blog_characteristics['Blog Tone']} audience.
-        3). Make sure your response content length is of {blog_characteristics['Blog Length']} words.
-        4). Include FAQs from 'People also Ask' section of provided context 'google search result'.
 
         I want the post to offer unique insights, relatable examples, and a fresh perspective on the topic.
         Here are some Google search results to spark your creativity on {search_keyword}:
@@ -51,6 +34,7 @@ def write_blog_google_serp(search_keyword, search_results):
         logger.error(f"Exit: Failed to get response from LLM: {err}")
         exit(1)
 
+
 def improve_blog_intro(blog_content, blog_intro):
     """Combine the given online research and gpt blog content"""
     prompt = f"""

lib/ai_writers/twitter_ai_writer.py

@@ -13,59 +13,84 @@ def tweet_writer():
         col1, col2 = st.columns([5, 5])
         with col1:
             hook = st.text_input(
-                    label="What's the tweet about:(Hook)",
+                label="**What's the tweet about? (Hook)**",
-                    placeholder="e.g., Discover the future of tech today!",
+                placeholder="e.g., Discover the future of tech today!",
-                    help="Provide a compelling opening statement or question to grab attention."
+                help="Provide a compelling opening statement or question to grab attention."
             )
 
         with col2:
-            # Collect user inputs with placeholders and help text
             target_audience = st.text_input(
-                label="Target Audience",
+                label="**Target Audience**",
                 placeholder="e.g., technology enthusiasts, travel lovers",
                 help="Describe the audience you want to target with this tweet."
             )
 
+        col3, col4 = st.columns([5, 5])
+        with col3:
+            tweet_tone = st.selectbox(
+                label="**Tweet Tone**",
+                options=["Humorous", "Informative", "Inspirational", "Serious", "Casual"],
+                help="Choose the tone you'd like the tweet to have."
+            )
+
+        with col4:
+            cta = st.text_input(
+                label="**Call to Action (Optional)**",
+                placeholder="e.g., Retweet this if you agree! (Leave blank if not applicable)",
+                help="Provide a call to action if you'd like to include one."
+            )
+
+        col5, col6 = st.columns([5, 5])
+        with col5:
+            keywords_hashtags = st.text_input(
+                label="**Keywords/Hashtags**",
+                placeholder="e.g., #AI #Innovation",
+                help="Provide 2-3 relevant keywords or hashtags."
+            )
+
+        with col6:
+            tweet_length = st.selectbox(
+                "Tweet Length (Optional)",
+                options=["Short (under 100 characters)", "Medium (100-200 characters)", "Long (200+ characters)"],
+                help="Choose the desired tweet length.",
+            )
+
     if st.button('**Write Tweets**'):
         if not target_audience or not hook:
             st.error("🚫 Please provide all required inputs.")
         else:
-            with st.status("Assigning AI professional to write your Google Ads copy..", expanded=True) as status:
+            with st.status("Assigning AI professional to write your tweets...", expanded=True) as status:
-                response = tweet_generator(target_audience, hook)
+                response = tweet_generator(target_audience, hook, tweet_tone, cta, keywords_hashtags, tweet_length)
                 if response:
                     st.subheader(f'**🧕👩: Your Tweets!**')
                     st.markdown(response)
                 else:
-                    st.error("💥**Failed to write Letter. Please try again!**")
+                    st.error("💥**Failed to generate tweets. Please try again!**")
 
 
-def tweet_generator(target_audience, hook):
+def tweet_generator(target_audience, hook, tone_style, cta, keywords_hashtags, tweet_length):
-    """ Email project_update_writer """
+    """ Tweet Generator """
 
     prompt = f"""
-    You are a social media expert creating tweets for an audience interested in {target_audience}. 
+	    You are a social media expert creating tweets for an audience interested in {target_audience}. 
-    Write 5 engaging, concise, and visually appealing tweets that each:
+	    Write 5 engaging, concise, and visually appealing tweets that each:
-
+	
-    1. Start with a compelling hook based on the following keywords: "{hook}"
+	    1. Start with a compelling hook based on the following input: "{hook}"
-    2. Include a compelling call to action.
+	    2. Include the following call to action: "{cta}" 
-    3. Use 2-3 relevant hashtags. 
+	    3. Use 2-3 relevant keywords/hashtags, including: "{keywords_hashtags}"
-    4. Adopt a tone that matches the following options: 
+	    4. Adopt the following tone/style: "{tone_style}"
-        - Humorous 
+	    5. Adhere to the following length requirement: {tweet_length} 
-        - Informative 
+	    
-        - Inspirational 
+	    Make sure to keep the tone consistent with the selected style and platform context.
-        - Serious 
+	
-        - Casual
+	    Here are some examples of call-to-actions to include (if no specific CTA was provided): 
-    5. Be under 100 characters (including spaces and punctuation). 
+	    - Retweet this if you agree!
-
+	    - Share your thoughts in the comments!
-    Here are some examples of call-to-actions to include:
+	    - Learn more at [link] 
-    - Retweet this if you agree!
+	    - Follow for more {target_audience} content.
-    - Share your thoughts in the comments!
+	
-    - Learn more at [link] 
+	    Output each tweet separated by a newline.
-    - Follow for more [topic] content
+	    """
-    - Like if you're excited about [topic]
-
-    Output each tweet separated by a newline.
-    """
 
     try:
         response = llm_text_gen(prompt)

lib/gpt_providers/text_generation/gemini_pro_text.py

@@ -2,7 +2,6 @@
 import os
 import sys
 from pathlib import Path
-import streamlit as st
 
 import google.generativeai as genai
 from dotenv import load_dotenv
@@ -21,7 +20,7 @@ from tenacity import (
 
 
 @retry(wait=wait_random_exponential(min=1, max=60), stop=stop_after_attempt(6))
-def gemini_text_response(prompt, temperature, top_p, n, max_tokens):
+def gemini_text_response(prompt, temperature, top_p, n, max_tokens, system_prompt):
     """ Common functiont to get response from gemini pro Text. """
     #FIXME: Include : https://github.com/google-gemini/cookbook/blob/main/quickstarts/rest/System_instructions_REST.ipynb
     try:
@@ -37,7 +36,9 @@ def gemini_text_response(prompt, temperature, top_p, n, max_tokens):
         "max_output_tokens": max_tokens,
     }
     # FIXME: Expose model_name in main_config
-    model = genai.GenerativeModel(model_name="gemini-1.5-pro-latest", generation_config=generation_config)
+    model = genai.GenerativeModel(model_name="gemini-1.5-pro-latest", 
+                                  generation_config=generation_config,
+                                  system_instruction=system_prompt)
     try:
         # text_response = []
         response = model.generate_content(prompt, stream=True)
@@ -45,7 +46,6 @@ def gemini_text_response(prompt, temperature, top_p, n, max_tokens):
             for chunk in response:
                 # text_response.append(chunk.text)
                 print(chunk.text)
-                #st.write(chunk.text)
         else:
             print(response)
         logger.info(f"Number of Token in Prompt Sent: {model.count_tokens(prompt)}")

lib/gpt_providers/text_generation/main_text_generation.py

@@ -27,7 +27,21 @@ def llm_text_gen(prompt):
         str: Generated text based on the prompt.
     """
     try:
+        # Read the config param to create system instruction for the LLM.
         gpt_provider, model, temperature, max_tokens, top_p, n, fp = read_return_config_section('llm_config')
+        blog_tone, blog_demographic, blog_type, blog_language, \
+            blog_output_format, blog_length = read_return_config_section('blog_characteristics')
+
+        # Construct the system prompt with the sidebar config params.
+        system_instructions = f"""
+        Below are the guidelines to follow:
+        1). You must respond in {blog_language} language.
+        2). Tone and Brand Alignment: Adjust your tone, voice, personality for {blog_tone} audience.
+        3). Make sure your response content length is of {blog_length} words.
+        4). The type of blog is {blog_type}, write accordingly.
+        5). The demographic for this content is {blog_demographic}.
+        6). Your response should be in {blog_output_format} format.
+        """
 
         #gpt_provider = check_gpt_provider(gpt_provider)
         # Check if API key is provided for the given gpt_provider
@@ -37,7 +51,7 @@ def llm_text_gen(prompt):
         if 'google' in gpt_provider.lower():
             try:
                 logger.info("Using Google Gemini Pro text generation model.")
-                response = gemini_text_response(prompt, temperature, top_p, n, max_tokens)
+                response = gemini_text_response(prompt, temperature, top_p, n, max_tokens, system_instructions)
                 return response
             except Exception as err:
                 logger.error(f"Failed to get response from gemini: {err}")
@@ -45,7 +59,7 @@ def llm_text_gen(prompt):
         elif 'openai' in gpt_provider.lower():
             try:
                 logger.info(f"Using OpenAI Model: {model} for text Generation.")
-                response = openai_chatgpt(prompt, model, temperature, max_tokens, top_p, n, fp)
+                response = openai_chatgpt(prompt, model, temperature, max_tokens, top_p, n, fp, system_instructions)
                 return response
             except Exception as err:
                 logger.error(f"Failed to get response from Openai: {err}")

lib/gpt_providers/text_generation/openai_text_gen.py

@@ -14,7 +14,7 @@ from tenacity import (
  
 
 @retry(wait=wait_random_exponential(min=1, max=60), stop=stop_after_attempt(6))
-def openai_chatgpt(prompt, model, temperature, max_tokens, top_p, n, fp):
+def openai_chatgpt(prompt, model, temperature, max_tokens, top_p, n, fp, system_prompt):
     """
     Wrapper function for OpenAI's ChatGPT completion.
 
@@ -45,7 +45,8 @@ def openai_chatgpt(prompt, model, temperature, max_tokens, top_p, n, fp):
         client = openai.OpenAI(api_key=os.getenv('OPENAI_API_KEY'))
         response = client.chat.completions.create(
             model=model,
-            messages=[{"role": "user", "content": prompt}],
+            messages=[{"role": "system", "content": system_prompt}, 
+                      {"role": "user", "content": prompt}],
             max_tokens=max_tokens,
             n=n,
             top_p=top_p,

lib/utils/alwrity_utils.py

@@ -131,8 +131,11 @@ def ai_seo_tools():
         "Generate OpenGraph Tags",
         "Optimize/Resize Image"
     ]
-    choice = st.selectbox("**👇Select AI SEO Tool:**", options, index=0, format_func=lambda x: f"📝 {x}")
+    
-
+    # Using st.radio instead of st.selectbox
+    choice = st.radio("**👇 Select AI SEO Tool:**", options, index=0, format_func=lambda x: f"📝 {x}")
+    
+    # Handle choices based on the selected option
     if choice == "Generate Structured Data - Rich Snippet":
         ai_structured_data()
     elif choice == "Generate Meta Description for SEO":

lib/workspace/alwrity_prompts/alwrity_system_instruction.prompts

@@ -0,0 +1,2 @@
+alwrity_system_instruction: |
+