WORK / DETAILS
Drift AI
Q4 2024ProductionAI

Drift AI

A retrieval-augmented generation layer that automated 60% of incoming support queries for Drift's team.

Overview

Drift's support team was spending the bulk of their day answering the same 40 questions. Good people doing work that shouldn't require people. We built a retrieval-augmented generation layer trained on their documentation, past tickets, and product changelogs. The system handles routine queries instantly, detects when something falls outside its confidence threshold, and routes those cases to a human agent — with full conversation context already attached. Support costs dropped. Response times improved. The team shifted their energy to the cases that actually needed them.

Highlights

RAG pipeline on internal docs and ticket history
Confidence-threshold escalation logic
Human handoff with full context preservation
OpenAI + Pinecone vector search
Support cost reduction from day one
Role
AI Engineer
Stack
Python // FastAPI // LangChain // Pinecone // OpenAI API
Result
60% of incoming support queries resolved automatically. Human agents freed up for high-complexity cases only.
Project Details
Slugdrift-ai
YearQ4 2024
StatusProduction
CategoryAI