Lexi - AI Language Interviewer

landing page
dashboard
conversation with ai agent
report modal

Inspiration

Recruiters are very familiar with an age-old issue: resumes often list language skills like “fluent” or “conversational,” but those claims are rarely verified and can be difficult to assess. We wanted to create an easy way for recruiters to confidently evaluate real language proficiency.

What it does

Lexi is an AI-powered language interview platform that verifies a candidate’s true language ability. Using our interface, recruiters easily create an AI interview-style assessment sent candidates. The interview consists of a short conversation and reading assessment with Lexi, who then returns an evaluation of fluency to the recruiter.

How we built it

Real-time voice pipeline: FastAPI backend with WebSockets streams audio in 500ms chunks. OpenAI Whisper handles multilingual speech-to-text, while GPT-4 generates adaptive follow-up questions based on the conversation context. Conversational UI: Next.js 14 App Router with a Google Meet-inspired interface. Voice activity detection shows when the candidate is speaking, and a live transcript sidebar displays the conversation in real-time. Session persistence: Supabase stores interview sessions, transcripts, and evaluations. Each interview gets a unique token-based URL—no login required for candidates. Evaluation engine: GPT-4 scores responses across fluency dimensions and generates a recruiter-friendly report with specific examples from the conversation.

Challenges we ran into

Rapid development led to merge conflicts and integration challenges. However, resolving them improved our collaboration workflow and strengthened our understanding of Git-based team development.

Accomplishments that we're proud of

We successfully built an system that conducts real conversational interviews and evaluates language proficiency automatically. This replaces traditional, slower, more manual processes.

What we learned

Integrating authentication and AI evaluation into a cohesive product.
Real-time audio streaming with WebSockets and voice activity detection
AI agent implementation for natural conversation flow
Full-stack development integrating OpenAI, Supabase, and SendGrid

What's next for Lexi - AI Language Interviewer

Video recording: Capture facial expressions and lip movement to assess communication confidence and non-verbal fluency cues
Custom pronunciation models: Train ML models on native speaker data to score accent accuracy and phoneme-level pronunciation
Deeper analytics — Track speech pace, filler word frequency, response latency, and vocabulary diversity over time to benchmark candidates against role-specific norms
Finally: To see Lexi become a widely-used, trusted partner for hiring managers worldwide.