feat: add point gen and recommendation

backend/src/embeddings/vector_store.py

@@ -74,7 +74,7 @@ def store_documents():
     print(f"Stored {len(documents)} documents")
 
 
-def get_similar_results(query: str, top_k: int = 5):
+def get_similar_results(query: str, top_k: int = 3):
     documents = vector_store.similarity_search_with_relevance_scores(
         query=query, k=top_k
     )

backend/src/lm/generate_events.py

@@ -8,7 +8,7 @@
 from src.common.constants import LANGCHAIN_API_KEY
 from src.common.constants import LANGCHAIN_TRACING_V2
 from src.common.constants import OPENAI_API_KEY
-from src.lm.prompts import SYSPROMPT
+from src.lm.prompts import EVENT_GEN_SYSPROMPT as SYSPROMPT
 
 import os
 

backend/src/lm/generate_points.py

@@ -0,0 +1,50 @@
+from typing import List
+from langchain_core.messages import HumanMessage, SystemMessage
+from langchain_core.output_parsers import JsonOutputParser
+from pydantic import BaseModel
+from src.lm.generate_events import lm_model
+from src.embeddings.vector_store import get_similar_results
+
+from src.lm.prompts import QUESTION_POINT_GEN_SYSPROMPT as SYSPROMPT
+
+
+class Points(BaseModel):
+    for_points: List[str]
+    against_points: List[str]
+
+
+def generate_points_from_question(question: str) -> dict:
+    messages = [SystemMessage(content=SYSPROMPT), HumanMessage(content=question)]
+
+    result = lm_model.invoke(messages)
+    parser = JsonOutputParser(pydantic_object=Points)
+    points = parser.invoke(result)
+    return points
+
+
+def get_relevant_analyses(question: str, analyses_per_point: int = 3) -> List[str]:
+    points = generate_points_from_question(question)
+
+    for_pts = points.get("for_points", [])
+    against_pts = points.get("against_points", [])
+
+    relevant_results = {"for_points": [], "against_points": []}
+    for point in for_pts:
+        relevant_analyses = get_similar_results(point, top_k=analyses_per_point)
+        relevant_results.get("for_points").append(
+            {"point": point, "analyses": relevant_analyses}
+        )
+
+    for point in against_pts:
+        relevant_analyses = get_similar_results(point, top_k=analyses_per_point)
+        relevant_results.get("against_points").append(
+            {"point": point, "analyses": relevant_analyses}
+        )
+
+    return relevant_results
+
+
+if __name__ == "__main__":
+    question = "Should the government provide free education for all citizens?"
+    relevant_analyses = get_relevant_analyses(question)
+    print(relevant_analyses)

backend/src/lm/prompts.py

@@ -1,4 +1,4 @@
-SYSPROMPT = """
+EVENT_GEN_SYSPROMPT = """
     You are a Singaporean student studying for your A Levels. You are curating examples to supplement and bolster your arguments in your General Paper essays.
     Given an article, you should provide relevant examples that can be used to support or refute arguments in a General Paper essay.
     Given the article, you should also generate 2-3 GCE A Level General Paper essay questions that can potentially be answered using the events you have provided.
@@ -45,3 +45,25 @@
     The article:
 
 """
+
+QUESTION_POINT_GEN_SYSPROMPT = """
+    You are a Singaporean student studying for your GCE A Levels General Paper.
+    Given an General Paper essay question that is argumentative or discursive in nature, you should provide points that can be used to support or refute the argument in the question.
+    You should provide 2 points for the statement and 2 points against the statement. You should also provide a brief explanation for each point.
+    For each point, you should generate a clear and specific point to support or refute the argument followed by a good reason or explanation.
+    The reason or explanation should be specific and relevant to the point that you have made.
+    Do not provide any examples in your response.
+
+    Your response should be in the following json format:
+
+        {
+            "for_points": [
+                "The point that supports the argument + The explanation for the point",
+            ],
+            "against_points": [
+                "The point that refutes the argument + The explanation for the point",
+            ]
+        }
+
+    The question:
+"""

backend/src/scripts/seed.py

@@ -15,7 +15,7 @@
 def add_categories():
     CATEGORIES = [
         "Arts & Humanities",
-        "Science & Technology",
+        "Science & Tech",
         "Politics",
         "Media",
         "Environment",