Meritus
Adaptive AI interview platform
Overview
Meritus is an adaptive AI interview platform that goes beyond a generic question bank — it tailors each interview session based on the candidate's resume, the role they're targeting, and how they respond as the session progresses. The goal is to give candidates something closer to a real technical interview: structured, personalized, and followed by feedback that actually reflects how they performed.
Technical Architecture & Specifications
Key Features & Implementation Highlights
- Adaptive session logic that adjusts interview flow based on candidate resume, target role, and prior responses
- Structured evaluation workflow producing recruiter-style feedback covering technical and behavioral performance
- Resilient session management capable of handling multiple concurrent interviews, including moderation safeguards and mid-session recovery
Challenges & Tradeoffs
The main engineering challenge was making interview sessions resilient to mid-session failures — network drops, server restarts, or browser crashes — while handling multiple concurrent interviews. Redis was chosen as the session state layer rather than persisting every interaction directly to Supabase, because interview sessions generate rapid, high-frequency state updates (each response triggers re-evaluation of the interview flow) that would overwhelm a relational database with write latency. Redis handles this volume with sub-millisecond writes, while Supabase stores the finalized session results. The tradeoff is that Redis is ephemeral by default, so session recovery required explicit checkpoint logic to persist enough state to resume a dropped session without replaying the entire interview.