How we taught AI to answer the hardest EV questions

A high volume of hard questions — where a wrong answer isn't an option.

Resolve EV's support team handled a high volume of complex EV technical inquiries, the kind of questions that take real expertise and real time to answer well, and where a wrong answer isn't an option.

A generic chatbot was never going to cut it. The answers customers needed lived in product documentation and years of support conversations — knowledge the team had, but that no system could search.

A support brain built from the knowledge they already had.

We combined an LLM and a vector database to answer highly technical questions. Documentation and years of email history were collected in a data lake, vectorized, and turned into searchable knowledge — and when the system isn't confident in an answer, it automatically escalates to a human expert instead of guessing.

Confidence matters as much as capability. Rather than letting the model guess, anything the system isn't sure about is escalated automatically — customers get an instant answer when the AI knows, and one from a real person when it doesn't.

An AI support system that answers like an expert — and knows its own limits.

• RAG support chatbotAnswers highly technical EV queries, grounded in documentation and years of support emails — not in guesswork.
• Data lake retrievalAtlas Vector Search over technical documentation and mailbox history makes the team's accumulated expertise searchable.
• Automatic human escalationUncertain answers are routed to human experts instead of being served to customers.
• Production AI stackLLM for reasoning, vector database for retrieval — proven, managed building blocks.

Expert answers in seconds, humans where they matter.

The chatbot now handles technical inquiries directly from the knowledge Resolve EV already had, while human experts focus on the cases that genuinely need them — support that scales without scaling the team.