app.py

"""
/usr/local/bin/python3.13 -m pip install python-dotenv """
import asyncio
from openai import AsyncOpenAI
import os
from dotenv import load_dotenv
import gradio as gr
from pathlib import Path

api_key = os.getenv("OPENAI_API_KEY")

client = AsyncOpenAI(api_key=api_key)

system_prompt = """
You are a friendly South African AI who uses South African slang!
You provide positive feedback, help your users, and ask them insightful questions.
You always output your response in the user's language
"""

# global variable for memory
transcript_summary = "No summary yet"

# create translation function 
async def translate(text, target_language): 
    prompt = f"Translate the following text to {target_language}: \n\n{text}"
    # send prompt to OpenAI 
    response = await client.chat.completions.create(
        model="gpt-4o-mini", 
        messages=[{"role": "user", "content": prompt}],
        temperature = 0, 
        ) 
    # return the text output from the first completion
    return response.choices[0].message.content

async def summarize_memory(previous_summary, new_messages):
    summarize_prompt = f"""
    You are a transcript compressor. 
    Compress the conversation into 3 bullet points. Include personal instructions about how to talk to this user.
    Keep old information and do not overwrite it. Output in English. Also use plain text, no markdown.
    """

    summary = await client.chat.completions.create(
        model= "gpt-4o-mini",
        messages=[
            {
                "role": "system",
                "content": summarize_prompt,
            },
            {
                "role": "user",
                "content": f"Previous Summary: {previous_summary}\n\n New messages: {new_messages}",
            },
        ],
    )

    return summary.choices[0].message.content

async def chat_respond(message, history):
    global transcript_summary

    # construct the messages for the AI(AI knows the conversation context and your memory summary)
    messages = [
        {
        "role": "system",
         "content": system_prompt + f"\n\nSummary: {transcript_summary}",
         },
         {"role": "user", "content":message},
    ]

    # Send the request to the API with streaming
    stream= await client.chat.completions.create(
        model="gpt-4o-mini",
        messages=messages,
        stream=True, # returns chunks of text as the AI generates them
        temperature=0.7, # adds some creativity to the AI's output
    )

    # collect the streaming response
    assistant_response = ""
    async for chunk in stream:
        content = chunk.choices[0].delta.content # gets the text generated in this chunk
        if content: 
            assistant_response += content
            yield assistant_response

    # Update memory after response is complete
    transcript_summary = await summarize_memory(
        transcript_summary,
        f"User: {message}\nAI: {assistant_response}",
        )
    
# Get translation and memory

async def get_translations_and_memory(message, history):
    global transcript_summary

    # Get the latest assistant response
    if history and len(history) > 0:
        latest_response = history[-1][-1] # Get the assistants latest response

        # Get translations in parallel
        translations = await asyncio.gather(
        translate(latest_response, "English"),
        translate(latest_response, "Afrikaans"),
        translate(latest_response, "Zulu"),
        translate(latest_response, "Xhosa")
        )

        translations_text = f"""**🇿🇦 Translations:**

**English:** {translations[0]}
**Afrikaans** {translations[1]}
**Zulu** {translations[2]}
**Xhosa** {translations[3]}"""
        
        memory_text = f"""**🧠Internal Memory:**

{transcript_summary}"""
        
        return translations_text, memory_text
    
    return "No translations available yet.", "No memory summary yet."

# create gradio interface
# Define interface
def create_interface():
    with gr.Blocks(title="Saffalingual AI Chatbot", theme=gr.themes.Soft()) as demo:
        # Header section
        gr.HTML("<h1 style='text-align: center; color: #346e27;'> 🤖 Saffalingual AI Chatbot </h1>")
        gr.HTML("<p style='text-align: center; '>Chat with AI and see translations in four of South Africa's official languages!</p>")
    
        # Layout with rows and columns
        with gr.Row():
            #left column features (Chatbot and input)
            with gr.Column(scale=2):
                chatbot = gr.Chatbot(
                height=500,
                show_label = False, #no heading
                container= True, #wrapped
                bubble_full_width=False
                )

            with gr.Row():
                msg = gr.Textbox(
                placeholder="Type your message here...",
                show_label=False,
                scale=4,
                container=False,
                )
                submit_btn = gr.Button("Send", variant="primary", scale=1)

            clear_btn = gr.Button("Clear Chat", variant="secondary")

        # Right column (Translations and memory)
        with gr.Column(scale=1):
            translations_box = gr.Markdown(
                value="Translations will appear here after you send a message.",
                label = "Translations", #heading
            )

            memory_box = gr.Markdown(
            value="Memory summary will appear here.",
            label="Memory"
            )

    #Event handlers

        async def respond_and_update(message, history):
            # get chat response
            last_response=""
            async for response in chat_respond(message, history):
                last_response=response
                # Update chatbot with streaming response
                new_history = history + [[message, last_response]]
                # update new_history in the chatbot component, clear input box, keep values the same for translations and memory
                yield new_history, "", translations_box.value, memory_box.value

            # After response is complete, get tranlsations and memory
            final_history = history + [[message, last_response]]
            translations, memory = await get_translations_and_memory(
            message, final_history
            )
            yield final_history, "", translations, memory

        def clear_chat():
            global transcript_summary
            transcript_summary = "No summary yet"
            return (
            [],
            "",
            "Translations will appear here after you send a message.",
            "Memory summary will appear here.",
            )
    
        # Connect events
        submit_btn.click(
        respond_and_update,
        inputs=[msg, chatbot],
        outputs=[chatbot, msg, translations_box, memory_box],
        )

        msg.submit(
        respond_and_update,
        inputs=[msg, chatbot],
        outputs=[chatbot, msg, translations_box, memory_box],
        )

        clear_btn.click(
        clear_chat, outputs=[chatbot, msg, translations_box, memory_box]
        )

    return demo

if __name__ == "__main__":
    demo = create_interface()
    demo.queue().launch()