Feature: Generate Opposite Perspective Node

new-backend/app/modules/langgraph_builder.py

@@ -8,13 +8,25 @@
     error_handler
     )
 
+from typing_extensions import TypedDict
+
+
+class MyState(TypedDict):
+    cleaned_text: str
+    facts: list[dict]
+    sentiment: str
+    perspective: str
+    score: int
+    retries: int
+    status: str
+
 
 def build_langgraph():
-    graph = StateGraph()
+    graph = StateGraph(MyState)
 
     graph.add_node(
                     "sentiment_analysis",
-                    sentiment.run_sentiment
+                    sentiment.run_sentiment_sdk
                     )
     graph.add_node(
                     "fact_checking",
@@ -34,21 +46,21 @@ def build_langgraph():
                     )
     graph.add_node(
                     "error_handler",
-                    error_handler
+                    error_handler.error_handler
                     )
 
     graph.set_entry_point(
-                    "sentiment_analysis"
+                    "sentiment_analysis",
                     )
 
-    graph.set_conditional_edges(
+    graph.add_conditional_edges(
         "sentiment_analysis",
         lambda x: (
             "error_handler" if x.get("status") == "error" else "fact_checking"
             )
     )
 
-    graph.set_conditional_edges(
+    graph.add_conditional_edges(
         "fact_checking",
         lambda x: (
             "error_handler"
@@ -57,7 +69,7 @@ def build_langgraph():
             )
     )
 
-    graph.set_conditional_edges(
+    graph.add_conditional_edges(
         "generate_perspective",
         lambda x: (
             "error_handler"
@@ -66,7 +78,7 @@ def build_langgraph():
             )
     )
 
-    graph.set_conditional_edges(
+    graph.add_conditional_edges(
         "judge_perspective",
         lambda state: (
             "error_handler"
@@ -80,12 +92,12 @@ def build_langgraph():
             else "store_and_send"
             )
     )
-    graph.set_conditional_edges(
+    graph.add_conditional_edges(
         "store_and_send",
         lambda x: (
             "error_handler"
             if x.get("status") == "error"
-            else None
+            else "__end__"
             )
     )
 

new-backend/app/modules/langgraph_nodes/generate_perspective.py

@@ -1,21 +1,27 @@
-from langchain.chains import LLMChain
-from langchain.prompts import PromptTemplate
+from app.utils.prompt_templates import generation_prompt
+from langchain_groq import ChatGroq
+from pydantic import BaseModel, Field
 
-prompt = PromptTemplate(
-    input_variables=["text", "facts"],
-    template="""Given the following article:
-{text}
 
-And the following verified facts:
-{facts}
+prompt = generation_prompt
 
-Generate a reasoned opposing perspective using chain-of-thought logic.
-"""
+
+class PerspectiveOutput(BaseModel):
+    reasoning: str = Field(..., description="Chain-of-thought reasoning steps")
+    perspective: str = Field(..., description="Generated opposite perspective")
+
+
+my_llm = "llama-3.3-70b-versatile"
+
+llm = ChatGroq(
+    model=my_llm,
+    temperature=0.7
 )
 
-my_llm = "groq llm"
+structured_llm = llm.with_structured_output(PerspectiveOutput)
+
 
-chain = LLMChain(prompt=prompt, llm=my_llm)
+chain = prompt | structured_llm
 
 
 def generate_perspective(state):
@@ -31,8 +37,15 @@ def generate_perspective(state):
         elif not facts:
             raise ValueError("Missing or empty 'facts' in state")
 
-        facts = "\n".join([f["snippet"] for f in state["facts"]])
-        result = chain.run({"text": text, "facts": facts})
+        facts_str = "\n".join([f"Claim: {f['original_claim']}\n"
+                               "Verdict: {f['verdict']}\nExplanation: "
+                               "{f['explanation']}" for f in state["facts"]])
+
+        result = chain.invoke({
+            "cleaned_article": text,
+            "facts": facts_str,
+            "sentiment": state.get("sentiment", "neutral")
+        })
     except Exception as e:
         print(f"some error occured in generate_perspective:{e}")
         return {

new-backend/app/modules/langgraph_nodes/judge.py

@@ -6,7 +6,7 @@ def judge_perspective(state):
         if not perspective:
             raise ValueError("Missing or empty 'perspective' in state")
 
-        score = 85 if "reasoned" in perspective else 40
+        score = 85 if "reasoning" in perspective else 40
     except Exception as e:
         print(f"some error occured in judge_perspetive:{e}")
         return {

new-backend/app/modules/langgraph_nodes/store_and_send.py

@@ -1,12 +1,13 @@
-from utils.vector_store import save_to_vector_db
+# from app.utils.vector_store import save_to_vector_db
 
 
 def store_and_send(state):
     # to store data in vector db
     try:
-        save_to_vector_db({
-            **state
-        })
+        print(state)
+        # save_to_vector_db({
+        #     **state
+        # })
     except Exception as e:
         print(f"some error occured in store_and_send:{e}")
         return {

new-backend/app/modules/pipeline.py

@@ -2,6 +2,7 @@
 from app.modules.scraper.cleaner import clean_extracted_text
 from app.modules.scraper.keywords import extract_keywords
 from app.modules.langgraph_builder import build_langgraph
+
 import json
 
 # Compile once when module loads

new-backend/app/utils/fact_check_utils.py

@@ -9,9 +9,9 @@ def run_fact_check_pipeline(state):
 
     result = run_claim_extractor_sdk(state)
 
+
     if state["status"] != "success":
         return result
-
     # Step 1: Extract claims
     raw_output = result["verifiable_claims"]
 
@@ -31,6 +31,7 @@ def run_fact_check_pipeline(state):
             print(f"❌ Search failed for: {claim} -> {e}")
         time.sleep(4)  # Add 4 second delay to prevent rate-limit
 
+
     if not search_results:
         return [], "All claim searches failed or returned no results."
 

new-backend/app/utils/prompt_templates.py

@@ -0,0 +1,32 @@
+from langchain.prompts import ChatPromptTemplate
+
+generation_prompt = ChatPromptTemplate.from_template("""
+You are an AI assistant that generates a well-reasoned '
+'counter-perspective to a given article.
+
+## Article:
+{cleaned_article}
+
+## Sentiment:
+{sentiment}
+
+## Verified Facts:
+{facts}
+
+---
+
+Generate a logical and respectful *opposite perspective* to the article.
+Use *step-by-step reasoning* and return your output in this JSON format:
+
+```json
+{{
+  "counter_perspective": "<your opposite point of view>",
+  "reasoning_steps": [
+    "<step 1>",
+    "<step 2>",
+    "<step 3>",
+    "...",
+    "<final reasoning>"
+  ]
+}}
+""")

new-backend/pyproject.toml

@@ -11,6 +11,8 @@ dependencies = [
     "fastapi>=0.115.12",
     "groq>=0.28.0",
     "langchain>=0.3.25",
+    "langchain-community>=0.3.25",
+    "langchain-groq>=0.3.2",
     "langgraph>=0.4.8",
     "logging>=0.4.9.6",
     "newspaper3k>=0.2.8",