Add application file

app.py

@@ -0,0 +1,46 @@
+# app.py
+import pickle, faiss, numpy as np
+from sentence_transformers import SentenceTransformer
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
+import gradio as gr
+
+# Load embeddings + index
+with open("faiss_index.pkl","rb") as f:
+    obj = pickle.load(f)
+index = obj["index"]
+texts = obj["texts"]
+metas = obj["metas"]
+
+embedder = SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")
+
+# Generator model (small/flan)
+TOKENIZER_MODEL = "google/flan-t5-small"
+tokenizer = AutoTokenizer.from_pretrained(TOKENIZER_MODEL)
+model = AutoModelForSeq2SeqLM.from_pretrained(TOKENIZER_MODEL)
+
+def retrieve(query, k=4):
+    q = embedder.encode([query], convert_to_numpy=True)
+    faiss.normalize_L2(q)
+    D, I = index.search(q, k)
+    results = []
+    for idx in I[0]:
+        results.append(texts[idx])
+    return results
+
+def generate_answer(query):
+    contexts = retrieve(query, k=4)
+    prompt = "Use the following context to answer the question.\n\nContext:\n" + "\n\n".join(contexts) + "\n\nQuestion: " + query + "\nAnswer:"
+    inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=1024)
+    out = model.generate(**inputs, max_length=256)
+    ans = tokenizer.decode(out[0], skip_special_tokens=True)
+    return ans
+
+with gr.Blocks() as demo:
+    gr.Markdown("# RAG FAQ Bot")
+    txt = gr.Textbox(lines=2, placeholder="Type question...")
+    btn = gr.Button("Ask")
+    out = gr.Textbox(lines=8)
+    btn.click(fn=lambda q: generate_answer(q), inputs=txt, outputs=out)
+
+if __name__ == "__main__":
+    demo.launch()

build_embeddings.py

@@ -0,0 +1,31 @@
+# build_embeddings.py
+from sentence_transformers import SentenceTransformer
+import os
+import faiss
+import pickle
+
+EMB_MODEL = "sentence-transformers/all-MiniLM-L6-v2"
+embedder = SentenceTransformer(EMB_MODEL)
+
+docs_dir = "docs"
+texts = []
+metas = []
+for fn in os.listdir(docs_dir):
+    if fn.endswith(".txt"):
+        txt = open(os.path.join(docs_dir, fn), encoding="utf-8").read()
+        # simple split by paragraph
+        chunks = [p.strip() for p in txt.split("\n\n") if p.strip()]
+        for i,c in enumerate(chunks):
+            texts.append(c)
+            metas.append({"source": fn, "chunk": i})
+
+embs = embedder.encode(texts, show_progress_bar=True, convert_to_numpy=True)
+d = embs.shape[1]
+index = faiss.IndexFlatIP(d)
+# normalize for cosine
+import numpy as np
+faiss.normalize_L2(embs)
+index.add(embs)
+with open("faiss_index.pkl", "wb") as f:
+    pickle.dump({"index": index, "texts": texts, "metas": metas}, f)
+print("built faiss_index.pkl")

requirements.txt

@@ -0,0 +1,5 @@
+gradio
+transformers
+sentence-transformers
+faiss-cpu
+torch