Update app.py

app.py

@@ -3,12 +3,12 @@ import json
 import requests
 import gradio as gr
 from bs4 import BeautifulSoup
-from textwrap import shorten
 
 from langchain_core.documents import Document
 from langchain_text_splitters import RecursiveCharacterTextSplitter
 from langchain_huggingface import HuggingFaceEmbeddings
 from langchain_community.vectorstores import Chroma
+from rank_bm25 import BM25Okapi  # <-- NEW Hybrid Search Import
 
 # ------------------ SCRAPE KUBERNETES DOCS ------------------ #
 
@@ -33,10 +33,7 @@ def scrape_page(name, url):
         if not content:
             return None
         text = content.get_text(separator="\n").strip()
-        return Document(
-            page_content=text,
-            metadata={"doc_id": name, "url": url}
-        )
+        return Document(page_content=text, metadata={"doc_id": name, "url": url})
     except:
         return None
 
@@ -59,7 +56,34 @@ retriever = vectordb.as_retriever(
     search_kwargs={"k": 5, "score_threshold": 0.4}
 )
 
-# ------------------ LLM CALL (OPENROUTER) ------------------ #
+# ------------------ HYBRID SEARCH ------------------ #
+
+bm25_corpus = [doc.page_content.split() for doc in chunks]
+bm25 = BM25Okapi(bm25_corpus)
+
+def hybrid_search(query, top_k=5):
+    # Vector Search
+    vector_results = retriever.invoke(query)
+
+    # BM25 Keyword Search
+    tokenized_query = query.lower().split()
+    bm25_scores = bm25.get_scores(tokenized_query)
+    bm25_ranked = sorted(zip(bm25_scores, chunks), key=lambda x: x[0], reverse=True)
+    bm25_results = [d for _, d in bm25_ranked[:top_k]]
+
+    # Combine + Remove duplicates
+    combined = vector_results + bm25_results
+    unique = []
+    seen = set()
+    for d in combined:
+        key = (d.metadata["doc_id"], d.page_content[:50])
+        if key not in seen:
+            seen.add(key)
+            unique.append(d)
+
+    return unique[:top_k]
+
+# ------------------ LLM CALL (OpenRouter) ------------------ #
 
 def call_llm(prompt):
     url = "https://openrouter.ai/api/v1/chat/completions"
@@ -72,19 +96,19 @@ def call_llm(prompt):
         "model": "meta-llama/llama-3.1-8b-instruct",
         "messages": [{"role": "user", "content": prompt}],
         "max_tokens": 400,
-        "temperature": 0.0,
+        "temperature": 0.0
     }
     r = requests.post(url, headers=headers, json=data)
     res = r.json()
     if "choices" in res:
         return res["choices"][0]["message"]["content"]
-    print("🚨 LLM ERROR:", res)
-    return "⚠️ Error: No response from model"
+    print("LLM ERROR:", res)
+    return "⚠️ Model failed. Please retry."
 
-# ------------------ BUILD ANSWER WITH CITATIONS ------------------ #
+# ------------------ RAG + CITATIONS ------------------ #
 
 def build_context_with_citations(query):
-    docs = retriever.invoke(query)
+    docs = hybrid_search(query)
     context = ""
     sources = []
     for i, d in enumerate(docs, start=1):
@@ -96,9 +120,9 @@ def build_context_with_citations(query):
 def answer_question(query, history):
     context, sources = build_context_with_citations(query)
     prompt = f"""
-Answer the question strictly using the context below.
-Every sentence must include citation like [1], [2].
-If missing info → reply: "Not in docs."
+Answer using ONLY the context below.
+Every sentence MUST include citations like [1], [2].
+If the answer is not in docs → respond "Not in docs."
 
 Question: {query}
 
@@ -106,8 +130,8 @@ Context:
 {context}
 """
     answer = call_llm(prompt)
-    src = "\n".join(sources) if sources else "No sources available."
-    history.append((query, answer + "\n\n---\nSources:\n" + src))
+    final = answer + "\n\n---\nSources:\n" + "\n".join(sources)
+    history.append((query, final))
     return history, ""
 
 # ------------------ GRADIO UI ------------------ #
@@ -115,22 +139,21 @@ Context:
 custom_css = """
 .source-box {
     background: #1e293b;
-    padding: 10px;
-    border-radius: 8px;
     color: #dbeafe;
+    padding: 10px;
+    border-radius: 7px;
     border: 1px solid #3b82f6;
 }
 """
 
 with gr.Blocks(theme="soft") as app:
     gr.HTML(f"<style>{custom_css}</style>")
-
     gr.HTML("<h1 style='text-align:center;color:#3b82f6'>☸ Kubernetes RAG Assistant</h1>"
-            "<p style='text-align:center;color:#cbd5e1'>Ask Kubernetes questions — answers include official docs citations 📌</p>")
+            "<p style='text-align:center;color:#cbd5e1'>Semantic + Hybrid Search • Official K8s Docs Cited 📌</p>")
 
     chat = gr.Chatbot(label="Conversation", height=450)
-    msg = gr.Textbox(label="Ask a question...", placeholder="What is a pod?")
-    clear = gr.Button("Clear Chat")
+    msg = gr.Textbox(label="Ask anything about Kubernetes…", placeholder="e.g., What is RBAC?")
+    clear = gr.Button("Clear Conversation")
 
     msg.submit(answer_question, [msg, chat], [chat, msg])
     clear.click(lambda: ([], ""), None, [chat, msg])