Alexis

Inspiration

87% of HR professionals report feeling overwhelmed and burned out, largely due to the demanding and time-intensive nature of recruiting. This strain often results in poor hiring decisions, mistakes that can cost companies up to $10,000 in lost productivity, turnover, and rehiring expenses. Statistics show that companies lose out on 5-20x of the employees salary for every bad hire. It’s common for a recruiter to become the bottleneck in the hiring process, since they’re often hiring for several roles at once and coordinating the resumes, interviews, offers, onboarding, contracts, etc… Particularly for early stage startups where the team is rapidly growing, each headcount can be incredibly valuable and they need fast and accurate hiring. One of our members experienced this issue first-hand: in two separate internships, she had to support and even own the hiring process due to HR overload. Any company, but particularly early stage startups, need a efficient recruiter, and that's what Alexis brings.

What it does

Alexis does end-to-end HR recruiting, helping users to do recruitment from job posting all the way to sending out offers. We do this by utilizing multi-agent frameworks which integrates to crucial platforms and make the workflow as seamless as possible.

How we built it

On the agentic side, we deploy to two agents: Agent S and Agent E. Agent S for now lives on Slack and responds in real-time to any messages mentioning it on Slack. Under the hood, Agent S utilizes Gemini API to understand context and give natural language response, when it’s necessary, it can call a bunch of tools integrated to emails, google calendar, and slack. To store persistent context, it utilizes Supabase to store messages. Agent E monitors the live inbox of whitelisted candidates and responds directly to any candidate question and it can also loop in a human when it deems necessary to do so to respond to uncommon questions or request. Both agents utilize Python and Flask to monitor messages and respond contextually.

On the app side, we utilized React + TypeScript on the frontend, Flask on the backend and Supabase for the database. Once a user uploaded a document, we store that using GCP and Google Buckets before sending to Supabase. Once, a user asked the AI agent, it can access these buckets.

Challenges we ran into

Authentication was a big problem especially since we were allowing an AI to access different services like Google Calendar and Slack. We spent a lot of time to setting up authentication across our devices and getting that to work in deployment. Tool calling configuration was also an issue where sometimes tools doesn’t work as expected and since it’s AI, it can be very hard to get replicated.

Accomplishments that we're proud of

We’re proud to try making a multi-agent app where users are able to automate a lot of HR tasks from end-to-end. This project involves a lot of backend and server integrations which we were happy to finish in only just 24 hours.

What we learned

We learn how to build multi-agent apps and utilized agentic frameworks to solve new problems. It was interesting how we could automate a lot of process and make it possible for one or two people to be empowered to do HR tasks than traditionally would take hundreds.