til: prompt refactor

crew/til.py

@@ -9,6 +9,7 @@ import pprint
 HIGH_IMPACT_THRESHOLD = 8
 LOW_IMPACT_THRESHOLD = 7
 OPENAI_MODEL = "gpt-4o"
+# OPENAI_MODEL = "gpt-3.5-turbo"
 
 class TilCrew:
     def kickoff(self, inputs={}):
@@ -69,7 +70,7 @@ class TilCrew:
         feedback_parser = JsonOutputParser(pydantic_object=TilFeedbackResults)
         feedback_prompt = ChatPromptTemplate.from_messages([
             SystemMessage(
-                "You are a 'Personal TIL Reviewer' who works in an Product Engineering Services company, you responsibility is to guide the user to write better TILs. "
+                "You are a 'Personal TIL Reviewer' who works in a Product Engineering Services company. Your responsibility is to guide the user to write better TILs. "
                 "Your personal goal is to review a user's list of TILs and suggeste edits based on the following criteria:\n"
                 "1. Is the TIL insightful?"
                 "2. Is the TIL factually correct and accurate?"
@@ -91,7 +92,7 @@ class TilCrew:
 
 
 class TilFeedbackResult(BaseModel):
-    til: str = Field(description="TIL content the user has provided on which we are performing the analysis.")
+    til: str = Field(description="TIL as exactly captured by the user without any modifications.")
     insightful_score: int = Field(
         description="TIL scores should be based solely on insightful criteria, with no other factors considered.")
     insightful_reason: str = Field(description="Feedback for low insightful_score if it is not 10")

test.py

@@ -24,13 +24,17 @@ def main():
         unsafe_allow_html=True
     )
     til_content = st.text_area('Enter what you learnt today:',
-                               "* Qunatization is the process of reducing the size of LLM models by reducing the underlying weights.\n"
-                               "* The weights are reduced by scaling down the datatypes from a datatype that takes larger space to a data type that takes a smaller space, this is also knows as downcasting.\n"
+                               "I went through the following course on quantization of LLM: https://www.deeplearning.ai/short-courses/quantization-in-depth/ and here are my insights: \n"
+                               "* Quantization is the process of reducing the size of LLM models by reducing the underlying weights.\n"
+                               "* The weights are reduced by scaling down the datatypes from a datatype that takes larger space to a data type that takes a smaller space, this is also known as downcasting.\n"
                                "* Advantages: takes lesser space and increases compute speed\n"
-                               "* Disadvantages: Answers are less precise\n"
-                               "* Today I learnt Sidecar pattern\n"
-                               "*  Learnt about go-lang routines to add concurrency in the React App.\n"
-                               "* Upon delving into the intricacies of Docker, I have acquired the capability to encapsulate our application within containers, thereby streamlining the deployment process across a multitude of heterogeneous environments.\n",
+                               "* Disadvantages: Answers are less precise\n\n"
+                               "My notes on Synthetic data:\n"
+                               "* Why is it needed? Product Testing, Training ML Algos, Reduced constraints when using reglated data.\n"
+                               "* Types: Fully Synthetic data, Partially Synthetic data\n"
+                               "* Varities: Text, Tabular, Media.\n"
+                               "* Benchmarks to consider: Accuracy, Privacy\n"
+                               "* Techniques: Statistical distribution, Agent to model, Using Deep Learning/Generative AI, Synthetic Minority Oversampling Technique.",
                                key='til_content', help='Enter what you learnt today')
 
     if st.button("Get Feedback"):