Talented AI

Find and recruit great talent effectively

Comment

Inspiration

I originally came to this hackathon to see if I could build an AI assistance for therapists. Great therapists are hard to come by. After talking to Arifa, we pivoted to recruiting. While I initially came to recruit great therapists, psychologists, and social workers, we decided on a more universally appealing platform to match talent with recruiters seamlessly, aided by the power of AI. The idea appealed to us as this platform could help companies recruit people, especially for filling tech roles based on remote video viewing. That is how talented AI was born.

What it does

We make it easy to do video screening of a large pool of candidates done by AI and capture the essence of the person, and allow the LLMs to match candidates with the key criteria required of a role. Our multimodal AI not only has a smooth onboarding experience for the interviewees but also is engaging and provides real-time feedback to the candidate and allows hiring managers to filter down very quickly and use the learning and the database for future recruitment efforts.

How we built it

Supabase, Lovable, Open AI, and, of course Eleven labs. We iterated through continuous prompting to enhance features such as resume builder, video analyzer, feedback summary, etc.

Challenges we ran into

Lovable works 80% of the time, but I had to spend time fixing issues when it made mistakes or got stuck in a loop, but it helped makeus to create something quickly in a hackthon time constraint.a

We encountered bugs with Eleven Labs not providing good error messages in their SDK, and an engineer had to assist by identifying the problem in the backend logs.

PostHog packages for LLM don't work on Deno, which is used by Supabase functions. So, I spent some time before giving up and exploring other options.

We still haven't figured out how to make AI more intelligent and more human. We did the best we could with prompt tuning in the time we had, but a lot more can be done to make this magical. For example, we could use WebSocket API for a chat interface between a recruiter and the candidate. We also wanted to use Deepgram to identify and prevent AI-generated voices from impersonating applicants.

There is no way to send context securely from the client side for Elevenlabs. We had to make the eleven labs key public, turn off security, and push all the info, including the system prompt while injecting context on a per-conversation basis. Ideally, the user should not be able to tweak the System prompt of the AI

Accomplishments that we're proud of

It works; we have video recording in the browser and interviews right there in the browser; no need to download anything; it is personalized.

What we learned

Both Elevenlabs API and lovable are great, Hackathons are energizing, and now we can use the tools I learned here for building something actually cool

What's next for Talented AI

We will open-source the code and then use this as a learning tool and pull it into the startup I am working on, which is solving mental health. Arifa plans to take talented AI further, as she plans to commercialize recruiting tech by combining it with learning and training AI agents for candidates. She was inspired by Vinter.me and wants to provide employment opportunities for the millions of engineering graduates looking for internships in India.

Built With

Share this project:

Updates