Brew | Devpost

logo
homepage

Inspiration

Having both gone through the recruiting process during our freshmen year, we both understand very well the challenges of getting an internship. Aside from just cold applying, we know the importance of reaching out to recruiters and industry professionals to not only gain insight but also build a strong professional network. Therefore, we made an autonomous cold outreach application to help make the experience a little less stressful.

What it does

Brew allows the user to enter a simple query about who they want to outreach and for what purpose, then it will generate a list of people who fit that query as well as personalized messages catered to their intentions. With just the press of a button, Brew will agentically send those messages via LinkedIn or email.

Furthermore, Brew provides user insights towards their cold messages. This will allow to user to monitor and track their outgoing messages. Brew also allows the user to discover those aligned to the user's fields of interests via the Explore page.

How we built it

We built Brew with the LINKD API, the Google Gemini LLM API, as well as Stagehand.

The LINKD API is used to fetch the appropriate list of LinkedIn Profiles according to the search query for the Brew page as well as fetching the people with aligned interests as the user in the Explore page.

Gemini is used to prompt engineer the query that goes into LINKD as well as generating the messages.

Stagehand is used for the agentic workflow. It is what allows the messages to be sent automatically both via LinkedIn and email.

Challenges we ran into

We faced some challenges in learning how to use stagehand, the agentic browser use framework. Specifically, it was difficult to get stagehand to remember authenticated states in Google and LinkedIn, which we needed to allow us to autonomously send messages via Gmail and LinkedIn.

Furthermore, it was tough getting the Gemini API's generated message to actually fit the user's intentions. Initially, we fed in too much of the user's data (even ones that weren't relevant) at once and the output wasn't desirable. We had to reduce the noise and specify the most important inputs in order to generate the catered and thoughtful message seen in the final build.

Accomplishments that we're proud of

We are very proud of the UIUX. It is a product of many rounds of design changes and discussions. However, we are both at the end very proud of the interface. We are also very proud of the agentic feature of Brew. Seeing the messages being typed in for the first time was thrilling and we are very happy with how it turned out.

What we learned

Aside from learning how to use 3 new great development tools (LINKD API, Gemini, and Stagehand), we also learned how to build fast and pivot as needed. In the future, we will be ready to meet tough deadlines and ship to users quickly.