AI Interview Agent

An interview is an open-ended conversation, but the value of one is in the structure you can pull back out: how a candidate reasoned, where they hesitated, what they actually demonstrated. This agent runs the conversation and produces that structure in real time.

Scoring a conversation as it happens

The system conducts a live interview and emits real-time structured scoring and feedback rather than a single verdict at the end. The hard part is reliability — turning free-form model output into a consistent, comparable schema — which I handle with structured output parsing so every response maps cleanly onto the scoring model.

Orchestrated, not prompted

The interview flow is orchestrated with LangGraph as explicit stages — ask, follow up, evaluate — over a FastAPI backend with a streaming channel for the live exchange. Treating the dialogue as a graph keeps it controllable instead of letting one long prompt drift.