Customer Service LLM Chatbot Program

I need a coder who already feels at home working with large-language-model frameworks to build a text-generation program that will power customer-service chatbot replies. The goal is straightforward: feed the bot a user query and receive a clear, helpful response that sounds natural, stays on-brand, and can be deployed inside our existing support interface.

Core needs
• Model handling: Need to use offline LLMs, preferably llama 3/opus and gemma as long as latency and accuracy stay competitive.
• Prompt/response logic: craft the prompt pipeline so it captures context (conversation history, customer metadata) and returns concise answers without hallucinating policies or numbers.
• Safety & tone controls: integrate guardrails for profanity, personally identifiable information, and brand voice consistency.
• Simple API wrapper: deliver clean Python code (FastAPI preferred) exposing a single endpoint: POST /reply with {user_message, conversation_id}.
• Quick-start docs: a README and sample cURL call are enough; I’ll plug it into the wider system from there.

Acceptance test
I’ll supply ten anonymized tickets. The program should answer at least eight of them with correct, policy-aligned replies in under two seconds each.

If you’ve already fine-tuned or deployed customer-service chatbots, let me know what stack you used and any metrics you hit. Looking forward to working together!