| import os | |
| import gradio as gr | |
| import requests | |
| from sentence_transformers import SentenceTransformer | |
| import chromadb | |
| HF_API_URL = os.getenv("HF_API_URL") | |
| HF_TOKEN = os.getenv("HF_TOKEN") | |
| EMB_MODEL = os.getenv("EMB_MODEL", "sentence-transformers/all-MiniLM-L6-v2") | |
| client = chromadb.Client() | |
| collection = client.get_or_create_collection("medini_memory") | |
| embedding_model = SentenceTransformer(EMB_MODEL, use_auth_token=HF_TOKEN) | |
| ROLES = [ | |
| "Tutor", "Product Manager", "Project Manager", "Designer/Developer", | |
| "Technical Writer", "Analyst", "Researcher", "Innovator", "Friend", "Video Generator" | |
| ] | |
| def retrieve_memory(user_input): | |
| results = collection.query(query_texts=[user_input], n_results=3) | |
| return " ".join(results.get("documents", [[]])[0]) | |
| def upsert_memory(user_input, response): | |
| emb = embedding_model.encode(user_input) | |
| collection.add(documents=[user_input], embeddings=[emb], ids=[str(hash(user_input))]) | |
| def medini_chat(user_input, role): | |
| context = retrieve_memory(user_input) | |
| payload = {"inputs": f"Role: {role}\nContext: {context}\nUser: {user_input}"} | |
| headers = {"Authorization": f"Bearer {HF_TOKEN}"} if HF_TOKEN else {} | |
| response = requests.post(HF_API_URL, headers=headers, json=payload) | |
| if response.status_code == 200: | |
| result = response.json() | |
| generated = result.get("generated_text", str(result)) | |
| upsert_memory(user_input, generated) | |
| return generated | |
| else: | |
| return f"Error: {response.text}" | |
| iface = gr.Interface( | |
| fn=medini_chat, | |
| inputs=[gr.Textbox(label="Your Message"), gr.Dropdown(ROLES, label="Select Role")], | |
| outputs="text", | |
| title="Medini - Multi-role AI Assistant", | |
| description="Chat with Medini as Tutor, PM, Designer, Analyst, Innovator, Friend, and more." | |
| ) | |
| if __name__ == "__main__": | |
| iface.launch(share=True) | |