Speak2Prep

Speak well to prepare. Don't just read/write here and there.

Comment

Inspiration

Our primary source of inspiration for Speak2Prep link are the challenges often faced by our team members while appearing for an interview, viva or any type of oral exam. We understand concepts, technical terms having decent knowledge & experience in specific technical domains. But when it comes to speaking/expressing our understanding to the other person either face-to-face/online, our brains tend to trip, which is also a common issue among many, as we know. So we think, why not share something useful with those who don't feel confident and clear before appearing for their next big moment!

What it does

This is a web based tool which helps aspiring candidates, students, professionals or anyone who want to feel well-prepared and improve their conceptual clarity before performing live in front of managers, founders, senior developers, professors, etc. The app hosts a live audio call with a customizable persona based agent. The agent initiates the conversation welcoming the user after user selects their topics & role. The agent quickly evaluates the answer in real-time, adapts the next question accordingly and provides a detailed scorecard at the end of the conversation. It also provides the user an improvement guide to fill the concept/topic gaps.

How we built it

We build it together by ideating the initial concept. We brainstormed, combining ideas, possible use cases and potential impact to create with our target audience of college students, recent graduates, tech. professionals. We decided to build the tool with AI-First-Critical-Human-Review (AFCHR) Development approach in which almost all the phases including core product development, debugging, code review (much of it) & architecture design has been undertaken by Gemini. With in-depth real-world testing, core business logic review & deployment being handled by humans.

Challenges we ran into

The major challenge we encountered was, how to implement speech-to-text for audio call especially considering real-time streaming capabilities. First we thought of using external speech-to-text API combining it with sockets to implement the call session. Even we started implementing it then we realize soon enough that it could mess a bit with the evaluation functionality in between the conversation and thus overall web socket connection. Then we got an idea of using multi-modal capable Gemini Live itself which would not only offer real-time audio streaming but native audio dialog flow with multi-modal capabilities if required.

Accomplishments that we're proud of

First of all, we're happy about that we have come up with something, build something useful and poised to make a potential impact as a team. Because it's our first-ever hackathon as a team. Individually our team leader has participated in a few before. And the other thing we like to mention that this tool can become a standard considering the vast scope and still increasing skill gap in today's world.

What we learned

We learned a lot and grind (quite a bit) as a team doing this project and will continue to learn and grow together, improving and scaling the app. We got a good understanding of Gemini Live and web sockets inside a serverless Next.js backend while implementing the app.

What's next for Speak2Prep

There's a lot more planned and currently in pipeline for "Speak2Prep". Such as call-session time options (5 min., 10 min., etc.), multiple topic upload options: external links, docs & images, quick QAs for different use cases, fun 1v1 friendly competitions (little later), leaderboards and more. Plus a well-designed minimal mobile app is in mid-term focus.

In short, "It's not a technical interview/oral exam preparation AI app we're creating, but a complete platform for those who need greater clarity with technical and communicative depth."

Built With

Share this project:

Updates