Create app.py

app.py

@@ -0,0 +1,119 @@
+import gradio as gr
+from transformers import AutoTokenizer, AutoModel, AutoModelForSeq2SeqLM, pipeline
+import torch
+import faiss
+import numpy as np
+
+# -------------------
+# CONFIG
+# -------------------
+DEVICE = "cpu"
+MAX_TOKENS = 256
+SEARCH_RESULTS = 3
+
+SYSTEM_MESSAGE = (
+    "You are a helpful medical tutor. Provide clear, accurate explanations "
+    "based strictly on the provided notes. If the answer isn't in the notes, "
+    "say you don't have enough information."
+)
+
+# -------------------
+# MODELS
+# -------------------
+# PubMedBERT for embeddings
+EMBED_MODEL_NAME = "microsoft/BiomedNLP-PubMedBERT-base-uncased"
+embed_tokenizer = AutoTokenizer.from_pretrained(EMBED_MODEL_NAME)
+embed_model = AutoModel.from_pretrained(EMBED_MODEL_NAME)
+
+# Flan-T5 for generation
+GEN_MODEL_NAME = "google/flan-t5-small"
+gen_tokenizer = AutoTokenizer.from_pretrained(GEN_MODEL_NAME)
+gen_model = AutoModelForSeq2SeqLM.from_pretrained(GEN_MODEL_NAME)
+generator = pipeline("text2text-generation", model=gen_model, tokenizer=gen_tokenizer, device=-1)
+
+# -------------------
+# SAMPLE NOTES
+# -------------------
+notes_dict = {
+    "Anatomy": """
+    The heart pumps blood through pulmonary and systemic circuits.
+    Oxygenated blood leaves the left ventricle and travels through the aorta.
+    """,
+    "Physiology": """
+    The respiratory system involves the exchange of oxygen and carbon dioxide
+    in the alveoli of the lungs.
+    """,
+    "Pharmacology": """
+    Paracetamol is used to relieve pain and reduce fever.
+    It acts centrally on the hypothalamic heat-regulating center.
+    """,
+}
+
+# -------------------
+# BUILD EMBEDDING INDEX
+# -------------------
+def embed_texts(texts):
+    inputs = embed_tokenizer(texts, padding=True, truncation=True, return_tensors="pt")
+    with torch.no_grad():
+        outputs = embed_model(**inputs)
+    embeddings = outputs.last_hidden_state.mean(dim=1).numpy().astype("float32")
+    return embeddings
+
+doc_map = []
+chunks = []
+for subject, content in notes_dict.items():
+    parts = [content[i:i + 400] for i in range(0, len(content), 400)]
+    for p in parts:
+        doc_map.append((subject, p))
+        chunks.append(p)
+
+embeddings = embed_texts(chunks)
+index = faiss.IndexFlatL2(embeddings.shape[1])
+index.add(embeddings)
+
+# -------------------
+# SEMANTIC SEARCH
+# -------------------
+def semantic_search(query, max_results=SEARCH_RESULTS):
+    query_emb = embed_texts([query])
+    distances, ids = index.search(query_emb, max_results)
+    results = []
+    for idx in ids[0]:
+        subject, chunk = doc_map[idx]
+        results.append((subject, chunk))
+    return results
+
+# -------------------
+# RESPONSE FUNCTION
+# -------------------
+def respond(message, history):
+    if not message.strip():
+        return "Please enter a question."
+
+    search_results = semantic_search(message)
+    context_parts = [f"From {s}:\n{t}" for s, t in search_results]
+    context_text = "\n\n".join(context_parts) if context_parts else "No relevant notes found."
+
+    prompt = f"{SYSTEM_MESSAGE}\n\nNotes:\n{context_text}\n\nQuestion: {message}\nAnswer:"
+
+    response = generator(prompt, max_new_tokens=MAX_TOKENS)[0]["generated_text"]
+    return response
+
+# -------------------
+# GRADIO APP
+# -------------------
+with gr.Blocks(title="PubMed Medical Tutor") as demo:
+    gr.Markdown("# 🧬 PubMed Medical Tutor (CPU Friendly)")
+    chatbot = gr.Chatbot(label="Tutor Chat")
+    msg = gr.Textbox(label="Ask a medical question...")
+    clear = gr.Button("Clear Chat")
+
+    def user_input(message, history):
+        bot_reply = respond(message, history)
+        history.append((message, bot_reply))
+        return "", history
+
+    msg.submit(user_input, [msg, chatbot], [msg, chatbot])
+    clear.click(lambda: None, None, chatbot, queue=False)
+
+demo.launch()