RAG adjustment

Author: lrsmello

happy-kids/api/migrations/0016_chat_memories_embedding.py

@@ -0,0 +1,19 @@
+# Generated by Django 5.1.5 on 2025-05-23 03:15
+
+import django.contrib.postgres.fields
+from django.db import migrations, models
+
+
+class Migration(migrations.Migration):
+
+    dependencies = [
+        ('api', '0015_chat_memories_user_message_sentiment'),
+    ]
+
+    operations = [
+        migrations.AddField(
+            model_name='chat_memories',
+            name='embedding',
+            field=django.contrib.postgres.fields.ArrayField(base_field=models.FloatField(), blank=True, null=True, size=None),
+        ),
+    ]

happy-kids/api/models.py

@@ -1,5 +1,6 @@
 from django.db import models
 from django.contrib.auth.models import User
+from django.contrib.postgres.fields import ArrayField
 
 class historicalSessions(models.Model):
     sessionDate = models.DateTimeField("session date", default="")
@@ -23,6 +24,7 @@ class chat_memories(models.Model):
     user_message_sentiment = models.CharField(max_length=20, blank=True, null=True) 
     chat_message = models.TextField()
     date_time = models.DateTimeField(auto_now_add=True)
+    embedding = ArrayField(models.FloatField(), blank=True, null=True) 
 
 class dim_question_type(models.Model):
     question_type = models.TextField(unique=True)

happy-kids/frontend/utils.py

@@ -5,6 +5,7 @@
 from datetime import timedelta
 from api import models
 import openai
+import numpy as np
 
 def get_short_term_memory(user_id):
     """Recupera a memória de curto prazo do banco de dados"""
@@ -61,3 +62,32 @@ def classify_sentiment(text):
     except Exception as e:
         print(f"Error during sentiment analysis: {e}")
         return None
+
+
+client = openai.OpenAI(api_key=os.getenv("OPENAI_API_KEY"))
+
+def get_embedding(text):
+    response = client.embeddings.create(
+        model="text-embedding-3-small",
+        input=text,
+    )
+    return response.data[0].embedding
+
+
+
+def cosine_similarity(a, b):
+    a = np.array(a)
+    b = np.array(b)
+    return np.dot(a, b) / (np.linalg.norm(a) * np.linalg.norm(b))
+
+def get_relevant_memories(user_id, question, top_k=3):
+    question_emb = get_embedding(question)
+
+    history = models.chat_memories.objects.filter(user_id=user_id).exclude(embedding=None)
+    scored = []
+    for h in history:
+        sim = cosine_similarity(question_emb, h.embedding)
+        scored.append((sim, h))
+    scored.sort(reverse=True, key=lambda x: x[0])
+    top_memories = [x[1] for x in scored[:top_k]]
+    return top_memories

happy-kids/frontend/views.py

@@ -1,4 +1,4 @@
-from .utils import get_short_term_memory, classify_sentiment
+from .utils import get_short_term_memory, classify_sentiment, get_embedding, get_relevant_memories
 from .forms import *
 from django.contrib.admin.views.decorators import staff_member_required
 from django.contrib.auth.decorators import login_required
@@ -16,7 +16,6 @@
 from django.db.models.functions import TruncDate
 from django.db.models import Count
 
-
 # Create your views here.
 @login_required
 @staff_member_required
@@ -89,6 +88,8 @@ def view_lulu(request):
 
         recent_memory = get_short_term_memory(user_id)
 
+        relevant_memories = get_relevant_memories(user_id, question, top_k=3)
+
         messages = [
             {
                 "role": "system",
@@ -100,6 +101,12 @@ def view_lulu(request):
             }
         ]
 
+        for mem in relevant_memories:
+            messages.append({"role": "user", "content": f"(Previously) {mem.user_message}"})
+            if mem.chat_message:
+                messages.append({"role": "assistant", "content": f"(Previously) {mem.chat_message}"})
+        
+
         messages.extend(recent_memory)
 
         messages.append({"role": "user", "content": question})
@@ -121,12 +128,14 @@ def stream_gpt():
             response_text_html = markdown.markdown(text=response_text, output_format='html')
 
             sentiment = classify_sentiment(question)
+            embedding = get_embedding(question)
 
             models.chat_memories.objects.create(
                 user_id=request.user.id,
                 user_message=question,
                 user_message_sentiment=sentiment,
-                chat_message=response_text
+                chat_message=response_text,
+                embedding=embedding
             )
 
             yield response_text_html 
@@ -137,7 +146,6 @@ def stream_gpt():
 
         return response_server
 
-
 ######################### Memos and Feedback #########################
 @login_required
 def view_save_memo(request):
@@ -186,7 +194,6 @@ def view_save_feedback(request):
         if not user_prompt or not bot_message or not feedback_type:
             return JsonResponse({"error": "Dados inválidos."}, status=400)
 
-        # Verifica se já existe feedback desse usuário para esse par prompt/resposta
         existing_feedback = models.chat_facts_feedback.objects.filter(
             user=request.user,
             user_prompt=user_prompt,
@@ -202,7 +209,6 @@ def view_save_feedback(request):
                 existing_feedback.save()
                 return JsonResponse({"status": "updated"})
 
-        # Nenhum feedback anterior → criar novo
         models.chat_facts_feedback.objects.create(
             user=request.user,
             user_prompt=user_prompt,