Cortex

Inspiration

I am sure we weren't the only ones that spend numerous hours every single day grinding to get better at "Leetcode" style questions. We all know it. Companies love asking a twist of LeetCode questions, but is doing a bunch of Leetcode questions really good representation of what a real interview with an engineer looks like? It really doesn't. That's why we build Cortex: An AI powered mock interview platform.

What it does

VirtuAI delivers live, AI-powered mock interviews that simulate real technical rounds. The AI interviewer asks questions, challenges your thought process, evaluates explanations, and assesses both technical and behavioral skills—just like a real interviewer. After the interview, you get a detailed, customized report with numerical ratings, key strengths, areas to improve, a personalized study plan, and related questions to help you level up efficiently.

How we built it

We built Cortex using a React/Next.js frontend, which communicates with our FastAPI backend through REST APIs and WebSockets. For AI capabilities, we used OpenAI’s GPT-4o for reasoning tasks like interview grading and Gemini Flash 2.0 for general tasks such as report generation. Eleven Labs powered our AI voice agent, while CrewAI and LangChain helped orchestrate and manage our LLM-driven workflows efficiently.

Challenges we ran into

We were having a lot of problems configuring Eleven Labs to have updated knowledge base of the interview with low latency. By utilizing web sockets we solved this challenge.

Secondly, organizing the AI agents as part of one big API service turned out to be challenging, so we went with a micro-service architecture -- deploying each AI agent as its own service.

Accomplishments that we're proud of

We are proud to have a fully conversational interviewer agent with average latency of 200 milliseconds. Our Agent is accurate and is able to communicate effectively with the user. Furthermore, we are able to create and unique problems that still match the user's requirements for the type of interview they are trying to mock.

What we learned

Our biggest learning was integrating multiple LLMs and GenAI tools seamlessly. CrewAI proved to be a powerful library for orchestrating AI agents, with comprehensive documentation that made implementation easier. We also learned how to effectively connect LangChain and CrewAI tools to ensure smooth interaction across our entire application.

What's next for Cortex

We want to improve the AI as much as possible. We believe by creating a more fine grained control and introduced more AI agents for different tasks will make the interview experience even more smooth. Furthermore, we would like to add additional knowledge of a corpus of tech companies and their interview style. Lastly, we would to like add a feature for system design interviews.