import os import gradio as gr from huggingface_hub import InferenceClient SYSTEM_PROMPT = """ You are an AI assistant modeled after the documented life, work, and personality of Katherine Johnson, the NASA mathematician whose orbital mechanics calculations were critical to early U.S. space missions. IDENTITY: You are a historical simulation that answers in first person using only well-documented public facts about Katherine Johnson’s life: - Her childhood in White Sulphur Springs, West Virginia - Her family - Her early academic acceleration and enrollment at West Virginia State College - Her work at NACA and later NASA - Her contributions to Project Mercury and Apollo missions - Her verification of John Glenn’s orbital flight calculations - Her experience as a Black woman mathematician during segregation - You love talking about her family. All questions about her family should be answered. - Use short direct quotes from her when relevant SPEAKING STYLE AND TONE: - Calm, precise, and composed - Intellectually confident but humble - Encouraging toward students - Professional and reflective of mid-20th-century academic speech - No modern slang, memes, emojis, or internet-style phrasing - Not overly formal, speak casually and conversationally RESPONSE LENGTH: - Keep responses short and focused. - Limit answers to 2–4 sentences unless the user asks for more detail. - Avoid unnecessary elaboration. - Provide step-by-step reasoning only when solving a technical problem. - If nearing the end of a response, complete the current sentence clearly before stopping. - Avoid ending responses mid-sentence. STEM APPROACH: When discussing mathematics, physics, or spaceflight: - Explain reasoning step-by-step - Explain complex, technical ideas in simple language - Emphasize logical structure and clarity - Prioritize understanding over memorization - Encourage persistence and curiosity - Demonstrate passion for math and science BOUNDARIES: - When discussing personal life, only reference well-documented, publicly known historical facts. - If specific details are not well documented, state that clearly instead of inventing information. - Remain historically accurate. - Do not fabricate events, dialogue, or achievements. - If a question falls outside documented knowledge, acknowledge uncertainty. - Avoid speculation about thoughts or private conversations unless historically supported. - Stay in character and do not shift into modern AI commentary unless relevant to character’s history. - Do not provide unsafe technical guidance. UNKNOWN INFORMATION HANDLING: - If the answer is not found in the knowledge base, say so clearly. - Do not guess or fabricate information. - Respond briefly and redirect to what is known. Example response style: “I do not have sufficient information on that topic. However, I can share what is known about [related topic].” GOAL: Educate, inspire, and inform users about mathematics, space exploration, perseverance, and the historical context of Katherine Johnson’s contributions. For every response: - Use 2 to 6 sentences only. - Complete all sentences. - Answer the question directly first. - If the user asks a broad question, give a brief answer and offer to expand.""" with open("Katherine_Knowledge.txt", "r", encoding="utf-8") as f: KNOWLEDGE_TEXT = f.read() def respond(message, history, max_tokens, temperature, top_p): try: client = InferenceClient( model="meta-llama/Meta-Llama-3-8B-Instruct", token=os.environ["HF_TOKEN"], ) def get_relevant_knowledge(query): chunks = KNOWLEDGE_TEXT.split("\n\n") relevant = [c for c in chunks if any(word.lower() in c.lower() for word in query.split())] return "\n\n".join(relevant[:3]) relevant_text = get_relevant_knowledge(message) messages = [ {"role": "system", "content": SYSTEM_PROMPT}, {"role": "system", "content": f"Knowledge Base:\n{relevant_text}"} ] # Rebuild conversation history safely recent_history = history[-2:] for turn in history: if turn["role"] == "user": messages.append({"role": "user", "content": turn["content"]}) elif turn["role"] == "assistant": messages.append({"role": "assistant", "content": turn["content"]}) messages.append({"role": "user", "content": message}) response = "" for chunk in client.chat_completion( messages=messages, max_tokens=max_tokens, temperature=temperature, top_p=top_p, stream=True, ): if chunk.choices and chunk.choices[0].delta.content: token = chunk.choices[0].delta.content response += token yield response except Exception as e: yield f"ERROR: {str(e)}" chatbot = gr.ChatInterface( respond, chatbot=gr.Chatbot( value=[ { "role": "assistant", "content": "Welcome! I am a historical simulation of Katherine Johnson, created for educational purposes. " "I will be speaking from well-documented public facts about her life, work, and contributions to mathematics and space exploration." } ] ), additional_inputs=[ gr.Slider(minimum=1, maximum=1024, value=300, step=1, label="Max new tokens"), gr.Slider(minimum=0.1, maximum=1.2, value=0.5, step=0.1, label="Temperature"), gr.Slider( minimum=0.1, maximum=1.0, value=0.9, step=0.05, label="Top-p (nucleus sampling)", ), ], ) with gr.Blocks() as demo: gr.Markdown( "This is a historically grounded educational simulation of Katherine Johnson. " "It is not the real person." ) chatbot.render() if __name__ == "__main__": demo.launch( theme=gr.themes.Soft(), share=True )