AI Business Development Website Chatbot and Voice Tool

I am looking for an experienced AI chatbot / LLM product developer to build a custom AI Business Development Director for my website. The project is focused on creating a polished website-based AI assistant that can speak with website visitors, qualify leads, explain the business, answer questions using approved company knowledge, and send structured summaries and actions to the team.

The tool should sit on a page of my website and should also appear as a popup when someone visits the site. You would need to do this. It should look and feel similar to ChatGPT or Claude, using their APIs or another suitable LLM API, but designed around our brand and business development use case. When a visitor opens the tool, they should see two options: “Write a message to me” and “Talk to me”. Before either conversation starts, the tool should request the visitor’s name, email address, job title and company.

Both the written chat and voice chat should connect to a selected LLM API. The chatbot will use approved business knowledge files and content, such as PDFs, PowerPoints, website links, service descriptions, training material, FAQs, commercial guardrails and business development instructions. The goal is for the chatbot to behave like an AI Business Development Director. It should be able to explain the business, answer questions, qualify leads, discuss potential opportunities, and handle commercial conversations within clear limits. It must not make unsupported claims, invent information, agree terms outside approved guardrails, or promise anything it is not authorised to promise. Where appropriate, it should escalate to a human.

For the voice version, I am open to the best recommended approach. This may involve OpenAI Realtime API, Whisper, text-to-speech, browser audio capture or another suitable voice stack. Please advise on what you recommend and why. Ideally, the system should also be able to remember returning visitors based on their email address, but this needs to be designed carefully and in a GDPR/privacy-aware way.

At the end of each conversation, or when the conversation is terminated, the tool should send an email to me and relevant colleagues with the visitor’s details, conversation summary, key discussion points, lead qualification, actions required and recommended follow-up. Ideally, the conversation should also be logged into Airtable or another CRM-style destination. The system should classify each lead, for example as high priority, medium priority, low priority or not relevant, and briefly explain why it has assigned that classification.

The tool needs to be integrated into my website and designed properly as a page/interface, not just dropped in as a basic widget. It should look professional, work smoothly on desktop, tablet and mobile, be mobile responsive and feel app-like on mobile. I also need a simple way to maintain and update the system after launch, including the ability to update knowledge files, FAQs, prompts, commercial guardrails, escalation emails, notification recipients and chatbot wording without needing to rebuild the whole tool.

The AI should be grounded in the approved knowledge base and should say when it does not know, rather than guessing or inventing information. It should also have clear human handoff rules, especially for pricing, legal terms, investment discussions, sensitive client matters or anything outside its approved knowledge base.
Please also include fallback behaviour, for example what happens if the LLM API fails, the voice system fails, the email notification fails or Airtable/CRM logging fails. I will need testing on desktop and mobile, as well as a short handover video showing how to update the knowledge base, prompts, settings and notification rules.

Please explain how you would build this, where it would sit technically, how it would be integrated into the website, how the chatbot and voice interface would work, how the knowledge base would be maintained and updated, how conversations would be stored or summarised, what subscriptions/APIs/hosting would be required, and what you can or cannot do.