video1.0<iframe src="https://www.loom.com/embed/e1ebc51fd7ac4dd1ac23efea3b58c2ee" frameborder="0" width="1920" height="1440" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>14401920Loomhttps://www.loom.com14401920https://cdn.loom.com/sessions/thumbnails/e1ebc51fd7ac4dd1ac23efea3b58c2ee-8dce62c02631f335.gif462.742This Loom presents Apoorv Gupta’s architecture for Variant, an AI lead qualification and first-response system with an API-first design. It normalizes inputs and runs two LLM steps to classify and respond behind one endpoint, with a full deployment plan that persists to a DB, syncs to a CRM, and uses a queue plus workers for scale. The implementation emphasizes guardrails such as JSON-only outputs that prohibit inventing pricing or features, and includes a live versus mock mode that returns an indicator of the path. He also covers handling garbage input, ambiguous leads, and provider failures with retries, fallback templates, and monitoring via logs, metrics, traces, and alerts for latency, error spikes, and bucket distribution shifts.