reafctors and providing context to tasks

README.MD

@@ -1,3 +1,10 @@
+---
+title: ArticleRecommender
+app_file: app.py
+sdk: streamlit
+sdk_version: 1.35.0
+---
+
 ### Initial Setup
 * Install python using conda by running the following command. `conda create -p venv python==3.11 -y`
 * Activate conda profile `conda activate ./venv`

agents/article_evaluator.py

@@ -1,6 +1,7 @@
 from crewai import Agent
 from llms.gpt import llm
 from tools.helpers import streamlit_callback
+from tools.scrape_website import scrape_tool
 
 article_evaluator = Agent(
     role="Recommended Article Evaluator",
@@ -16,5 +17,6 @@ article_evaluator = Agent(
     ),
     allow_delegation=False,
     llm=llm,
+    tools=[scrape_tool],
     step_callback=streamlit_callback
 )

app.py

@@ -1,16 +1,16 @@
+from dotenv import load_dotenv
+load_dotenv()
+
 import json
 import streamlit as st
 
 import utils.settings as settings
 
 from crew.article_suggestion import article_recommendation_crew
-from dotenv import load_dotenv
 from utils.write_to_json import write_dict_to_json as write_dict_to_json
 
 settings.init()
 
-load_dotenv()
-
 
 def icon(emoji: str):
     """Shows an emoji as a Notion-style page icon."""
@@ -41,12 +41,18 @@ def main():
                 "Tool Use: The LLM is given tools such as web search, code execution, or any other function to help it gather information, take action, or process data. "
                 "Planning: The LLM comes up with, and executes, a multistep plan to achieve a goal "
                 "Multi-agent collaboration: More than one AI agent work together, splitting up tasks and discussing and debating ideas, to come up with better solutions than a single agent would.\n\n"
-                # "GenAI Multi-Agent Systems (https://thenewstack.io/genai-multi-agent-systems-a-secret-weapon-for-tech-teams/)\n"
+                "GenAI Multi-Agent Systems (https://thenewstack.io/genai-multi-agent-systems-a-secret-weapon-for-tech-teams/)\n"
                 "Multi-agent systems go beyond the task-oriented roles to truly super-charge development and strategy teams. "
                 "Successful multi-agent systems act as a “digital twin” for your development team. "
                 "Different Approaches: 1. Centralized, with one agent in the center that collects and assimilates all the other outputs. "
                 "2. Distributed, where there is no central controller and the agents coordinate directly with one another in an “agent swarm. "
-                "3. Hierarchical, where agents are organized in teams or hierarchical layers.\n",
+                "3. Hierarchical, where agents are organized in teams or hierarchical layers.\n\n"
+                "LLM Model Quantisation\n"
+                "Different Methods for Compression: Pruning, Knowledge Distiallation and Quantization."
+                "Quantization process represents the model weights in lower precession which is also known as downcasting."
+                "Quanitzatoin Error is the difference in the weights of the quantized model and the original model."
+                "Advantages of Quanitzation: Reduced memory footprint, increased compute and speed of inferrence."
+                "Disadvantages of Quantization: Less precise.\n\n",
                 height=400,
             )
             st.markdown("")  # Adding a blank space here
@@ -71,8 +77,9 @@ def main():
 
         st.subheader("", anchor=False, divider="rainbow")
 
-        json_data = json.loads(result)
-        articles_list = json_data.get("articles", None)
+        # json_data = json.loads(result)
+        # articles_list = json_data.get("articles", None)
+        articles_list = settings.articles.values()
 
         if articles_list is None:
             st.markdown("No articles found.")
@@ -81,7 +88,10 @@ def main():
             for article in articles_list:
                 st.markdown(f"# {article['title']}")
                 st.markdown(f"**URL:** [{article['url']}]({article['url']})")
-                st.markdown(f"**Pitch:** {article['pitch']}")
+                st.markdown(f"**Pitch:** {article.get('pitch', '')}")
+                st.markdown(f"**Evaluation Score:** {article.get('evaluation_score', '')}")
+                st.markdown(f"**Evaluation Reason:** {article.get('evaluation_reason', '')}")
+                st.markdown(f"**Reason For Recommendation:** {article.get('reason_for_recommendation', '')}")
                 st.markdown("---")
 
 

crew/article_suggestion.py

@@ -15,7 +15,7 @@ article_recommendation_crew = Crew(
     tasks=[learning_profile_task, article_suggestion_task,
            evaluation_task, article_pitch_task],
     verbose=True,
-    memory=True,
+    # memory=True,
     manager_llm=llm,
     embedder={
         "provider": "openai",
@@ -23,5 +23,4 @@ article_recommendation_crew = Crew(
                     "model": 'text-embedding-3-small'
         }
     }
-    # manager_llm=gemini_client
 )

requirements.txt

@@ -4,3 +4,4 @@ crewai_tools
 langchain_community
 langchain_google_genai
 langchain_openai
+streamlit

tasks/create_article_pitch.py

@@ -10,6 +10,8 @@ from typing import List
 
 from agents.curiosity_catalyst import curiosity_catalyst
 from tools.scrape_website import scrape_tool
+from tasks.create_learning_profile import learning_profile_task
+from tasks.evaluate_articles import evaluation_task
 
 
 class PitchedArticle(BaseModel):
@@ -23,23 +25,23 @@ class PitchedArticles(BaseModel):
 
 
 def callback_function(output: TaskOutput):
-    evaluated_articles = json.loads(output.exported_output)['articles']
+    evaluated_articles = json.loads(output.exported_output)
     st.markdown(evaluated_articles)
-    # for article in evaluated_articles:
-    #     settings.articles[article['url']]['pitch'] = article['pitch']
+    for article in evaluated_articles['articles']:
+        settings.articles[article['url']]['pitch'] = article['pitch']
     st.markdown("### Create Article Pitch is executed successfully!")
 
 
 article_pitch_task = Task(
     description=(
-        "Create a pitch only for the articles that have passed evaluation and no other links. "
+        "Create a pitch only for the articles that have been evaluated and no other links. "
         "Craft the pitch so to that it teases the article's most intriguing aspects, "
         "by posing questions that the article might answer or "
         "highlighting surprising facts to pique the user's curiosity "
         " to read the article for incremental learning."
     ),
     expected_output=(
-        "List of all the artilces that have passed evaluation phase along with their url and pitch statement."
+        "List of all the artilces that have been evaluated phase along with their url and pitch statement and no other new urls."
     ),
     output_json=PitchedArticles,
     output_file="pitched_articles.json",
@@ -47,4 +49,5 @@ article_pitch_task = Task(
     agent=curiosity_catalyst,
     async_execution=False,
     callback=callback_function,
+    context=[learning_profile_task, evaluation_task]
 )

tasks/evaluate_articles.py

@@ -9,6 +9,8 @@ from pydantic import BaseModel
 from typing import List
 
 from agents.article_evaluator import article_evaluator
+from tasks.create_learning_profile import learning_profile_task
+from tasks.new_article_suggestion import article_suggestion_task
 
 
 class EvaluatedArticle(BaseModel):
@@ -40,14 +42,16 @@ evaluation_task = Task(
         "Score the articles on the scale of 1 to 10, "
         "1 being doesn't provide incremental learning and "
         "10 being provides incremental learning to the user."
+        "Evaluate only articles that have been suggested to the user and no other articles."
     ),
     expected_output=(
-        "List of article titles along with their links, evaluation score and "
-        "evaluation reason w.r.t to insights captured by the user."
+        "List of article titles with their URLs, evaluation scores, "
+        "and evaluation reasons w.r.t insights captured by the user."
     ),
     output_json=EvaluatedArticles,
     output_file="evaluated_articles.json",
     agent=article_evaluator,
     async_execution=False,
     callback=callback_function,
+    context=[learning_profile_task, article_suggestion_task]
 )