feat: colocando o output do resumo como json

langchain_backend/main.py

@@ -1,5 +1,5 @@
 import os
-from langchain_backend.utils import create_prompt_llm_chain, create_retriever, getPDF, create_llm
+from langchain_backend.utils import create_prompt_llm_chain, create_retriever, getPDF, create_llm, create_prompt_llm_chain_summary
 from langchain_backend import utils
 from langchain.chains import create_retrieval_chain
 from langchain_huggingface import HuggingFaceEmbeddings
@@ -43,7 +43,7 @@ def get_llm_answer_summary(system_prompt, user_prompt, pdf_url, model, isIterati
   print('\n\n\n')
   pages = getPDF(pdf_url)
   if not isIterativeRefinement:
-    rag_chain = create_prompt_llm_chain(system_prompt, model)
+    rag_chain = create_prompt_llm_chain_summary(system_prompt, model)
 
     results = rag_chain.invoke({"input": user_prompt, "context": pages})
 

langchain_backend/utils.py

@@ -8,6 +8,9 @@ from langchain_core.prompts import ChatPromptTemplate
 from langchain_huggingface import HuggingFaceEndpoint, HuggingFaceEmbeddings
 from setup.environment import default_model
 from uuid import uuid4
+from langchain_core.output_parsers import JsonOutputParser
+from langchain_core.pydantic_v1 import BaseModel, Field
+from typing import List
 
 os.environ["LANGCHAIN_TRACING_V2"]="true"
 os.environ["LANGCHAIN_ENDPOINT"]="https://api.smith.langchain.com"
@@ -78,17 +81,37 @@ def create_llm(modelParam):
     )
 
 
+class Resumo(BaseModel):
+    nome_do_memorial: str = Field()
+    argumentos: str = Field()
+    jurisprudencia: str = Field()
+    doutrina: str = Field()
+    palavras_chave: List[str] = Field()
+
+def create_prompt_llm_chain_summary(system_prompt, modelParam):
+  model = create_llm(modelParam)
+  
+  
+
+  system_prompt = system_prompt + "\n\n" + "{context}"
+  prompt = ChatPromptTemplate.from_messages(
+      [
+          ("system", system_prompt),
+          ("human", "{input}"),
+      ]
+  )
+  question_answer_chain = create_stuff_documents_chain(model, prompt)
+  final_chain = question_answer_chain | JsonOutputParser(pydantic_object=Resumo)
+  return final_chain
+
+
 DEFAULT_SYSTEM_PROMPT = """
 
 You are a highly knowledgeable legal assistant specializing in case summarization. Your task is to provide comprehensive and accurate summaries of legal cases while maintaining a professional and objective demeanor. Always approach each case with careful consideration and analytical rigor.
 
 First, you will be given a document to analyze:
 
-Next, you will receive a specific request for summarization:
-
-<summary_request>
-{{resuma esse memorial}}
-</summary_request>
+Next, you will summarize a content provided.
 
 Before providing your summary, follow these steps:
 
@@ -102,18 +125,9 @@ Before providing your summary, follow these steps:
 
 3. Maximal Marginal Relevance: Apply the principles of Maximal Marginal Relevance to ensure your summary includes diverse, relevant information while avoiding redundancy. Prioritize information that is both relevant to the summary request and adds new insights not already covered.
 
-After completing these steps, provide your summary in the following format:
-
-<summary>
-{
- "nome_do_memorial": "",
- "argumentos": "",
- "jurisprudencia": "",
- "doutrina": "",
- "palavras_chave": [
- ]
-}
-</summary>
+After completing these steps, provide your summary of around 5000 characteres in a JSON format with the keys and types: nome_do_memorial (string), argumentos(string), jurisprudencia(string), doutrina(string), palavras_chave(array of strings).
+
+Please, make the format of the summary in BBcode and use as much as possible lists in BBcode format
 
 Remember:
 - Always prioritize relevance to the summary request.