Update app.py

app.py

@@ -1,17 +1,22 @@
-import gradio as gr
-from transformers import pipeline
-
-# Load a base model for Q&A (can swap with your fine-tuned one later)
-qa_pipeline = pipeline("question-answering", model="distilbert-base-cased-distilled-squad")
-
-# Simple FAQ memory (can expand with Koda-specific Q&A)
-FAQ = {
-    "delivery": "We deliver appliances within 3–5 business days after order confirmation.",
-    "billing": "Billing occurs monthly on your subscription start date.",
-    "returns": "You may return appliances within 30 days for a full refund.",
-    "washer noise": "This is often caused by an unbalanced load. Try redistributing clothes evenly.",
-    "fridge warm": "Check if the vents are blocked or if the temperature setting is correct.",
-}
+import pandas as pd
+from sentence_transformers import SentenceTransformer
+from sklearn.metrics.pairwise import cosine_similarity
+import numpy as np
+
+faq_df = pd.read_csv("data/koda_faq.csv")
+faq_questions = faq_df["question"].tolist()
+faq_answers = faq_df["answer"].tolist()
+
+embedder = SentenceTransformer("all-MiniLM-L6-v2")
+faq_embs = embedder.encode(faq_questions, normalize_embeddings=True)
+
+def retrieve_answer(user_q, top_k=1, thresh=0.35):
+    q_emb = embedder.encode([user_q], normalize_embeddings=True)
+    sims = cosine_similarity(q_emb, faq_embs)[0]
+    idx = int(np.argmax(sims))
+    if sims[idx] >= thresh:
+        return faq_answers[idx], float(sims[idx])
+    return None, float(sims[idx])
 
 def koda_assistant(user_input, history):
     # Check if input matches FAQ keywords