IntroCall is an AI-powered interview screening web platform we built at NexaurAI. It records live interviews, transcribes speech in real time, evaluates candidate responses using Gemini AI, and outputs a ranked scorecard — all automatically. This article covers our journey: what we built, what broke, and what worked.
Most companies waste hundreds of hours reviewing resumes and conducting first-round interviews that could be automated. A recruiter at a mid-size company might spend 4–6 hours per day on initial phone screens — asking the same 10 questions, manually taking notes, and then trying to compare scores across candidates days later.
We wanted to fix this. Not by removing humans from hiring, but by handling the repetitive, structured first layer — so hiring teams can focus their energy on the candidates who actually matter.
At its core, IntroCall is a web-based interview platform with three main capabilities:
We chose Next.js because it gave us a robust, server-side-rendered web platform that works seamlessly across all browsers and devices. Its API routes simplified our backend logic, and the built-in optimisation features ensured fast load times even with heavy media content. The component-based architecture also allowed our team to iterate rapidly on the recruiter dashboard and candidate portal independently.
We evaluated GPT-4o, Claude Sonnet, and Gemini 1.5 Pro for the scoring engine. Gemini won on consistency — its structured output (JSON mode) was more reliable, and its context window allowed us to pass the full transcript plus job description in a single prompt. We used a structured prompt that returned scores across 5 dimensions.
Our API layer is Node.js on Google Cloud Run, with Firebase for auth, Firestore for data, and Firebase Storage for audio/video files. This stack allowed rapid iteration without managing infrastructure. Cold start times were our main concern, but Cloud Run min instances kept latency acceptable.
Building IntroCall taught us several hard lessons:
Recruiters using IntroCall report spending less than 20 minutes reviewing an initial cohort of candidates — down from 3–4 hours. The AI summaries are accurate enough that they trust the ranking without needing to re-listen to every recording.
The biggest lesson from IntroCall wasn't technical — it was strategic. We spent the first 6 weeks building features that users didn't care about. If we had done proper market research and user interviews before starting development, we would have prioritised the recruiter dashboard over the candidate-facing portal. The candidates didn't choose us — the recruiters did.
This is exactly why at NexaurAI, we now offer App Idea Validation as a standalone service before any development begins. We don't want our clients to make the same mistake we did.
IntroCall is live and available for teams looking to automate their first-round interview process. You can request a demo below or explore the platform directly.