AI Customer Support Chatbot
Budget: £1,500 – £3,000 GBP
I need a production-ready chat agent that can handle frontline customer support in real time, generating dynamic answers instead of fixed scripts. The bot must:
• Draw live information from my customer database, product catalog, support articles, and a separate set of internal training documents.
• Create quizes, or downloadable PDFs in the users chosen language
• Interpret a user’s intent, pull the correct data, and reply conversationally in clear English.
• Maintain context across a session, hand off to a human seamlessly when confidence drops, and log the full conversation for analytics.
You are free to build with Dialogflow, Rasa, LangChain, OpenAI, or a comparable NLP framework as long as you can deploy it to our existing platform (JavaScript - SEEDL) and expose a REST hook for our website. Users will access this predominantly on a desktop device.
Acceptance criteria
• Accurate answers for at least 90 % of the top 100 support queries we will supply.
• Response time under two seconds on a standard AWS t3.medium instance.
• Secure read-only access to the data sources; no PII may be stored in the model.
• A brief hand-over flow that triggers when confidence < 0.6, including a transcript for the live agent.
Deliverables
• Source code with setup scripts
• Trained model or prompt templates
• Deployment guide & one-hour walkthrough
• One week of post-launch bug support
• Draw live information from my customer database, product catalog, support articles, and a separate set of internal training documents.
• Create quizes, or downloadable PDFs in the users chosen language
• Interpret a user’s intent, pull the correct data, and reply conversationally in clear English.
• Maintain context across a session, hand off to a human seamlessly when confidence drops, and log the full conversation for analytics.
You are free to build with Dialogflow, Rasa, LangChain, OpenAI, or a comparable NLP framework as long as you can deploy it to our existing platform (JavaScript - SEEDL) and expose a REST hook for our website. Users will access this predominantly on a desktop device.
Acceptance criteria
• Accurate answers for at least 90 % of the top 100 support queries we will supply.
• Response time under two seconds on a standard AWS t3.medium instance.
• Secure read-only access to the data sources; no PII may be stored in the model.
• A brief hand-over flow that triggers when confidence < 0.6, including a transcript for the live agent.
Deliverables
• Source code with setup scripts
• Trained model or prompt templates
• Deployment guide & one-hour walkthrough
• One week of post-launch bug support