Luup | Devpost

Luup - Making Networking Easier

Inspiration

Our team consists of CS students who struggled to get referrals and make connections in the industry. LinkedIn's premium tier is the current best solution; however, it was unaffordable for us and many of our college peers. The whole process is a struggle, how do you find people to network with? And if you do, how do you draft up a personalized message every single time? As if that wasn't enough...how do you track all of that?? Feeling this pain point ourselves, we realized that using Generative AI can automate much of the grunt work involved and decided to build Luup, a tool that makes networking easier.

What it does

Luup has 3 simple parts, people search, message drafting, and an outreach dashboard.

People Search

People search does exactly what it sounds like it does—searches for people for you to reach out to. Once a user puts in the role they're looking for, company they want to work at, and university they go to, Luup searches for the best matches for them to reach out to.

Message Drafting

Once the user selects someone to message, this feature takes in information from the user's LinkedIn, and the LinkedIn of who they want to reach out to, and drafts up a personalized email and LinkedIn DM that they can use to request a coffee chat.

Outreach Dashboard

After the user sends an email/DM, the Dashboard will automatically be updated to track who the user has reached out to and if they have been responded to. This allows the user to see everyone they've reached out to and what companies they can get referrals to.

How we built it

We built the front end using Next.js for its simple routing capabilities and Tailwind for styling. The back end is built using Firebase for storage and Flask for API development. The email and LinkedIn dm generator uses Google's Gemini API. It uses a fine-tuned model trained on synthetic data produced by structured prompts via Gemini.

Challenges we ran into

We ran into many challenges throughout the competition, but that's what makes hackathons fun! We first struggled with how to get data about professionals and tried many different APIs and databases till we found Google's Custom Search API which was perfect for this use case. Another struggle was making the messages not sound obviously written by AI. Ideally, we could’ve fine-tuned a model off a massive set of handwritten outreach messages, but this was difficult to find or hard to access. So we had to get creative and ended up fine-tuning the model off synthetic data produced by structured prompting. Another issue that we ran into was time. We simply did not have enough time to complete all the features that we initially wanted to implement when we first thought of this idea for the hackathon. We were initially extremely ambitious but later learned that it was just not realistic to get it all done before the deadline. Because of this, we had to put ourselves into the user’s shoes and think about which features were the most important to build first—this helped us prioritize all of our ideas in a way that allowed us to build a tool that still addresses the problems despite not including absolutely everything we wanted.

Accomplishments that we're proud of

We are proud of the smaller details that help our product address the pain points on the journey of college students trying to network, schedule coffee chats, and land referrals. Small things like automatically getting data from the user’s LinkedIn profile, and tracking all their connections to reference later for applications, make Luup address some of the most common issues with the whole process. We also have the generated messages incorporate industry and college-specific humor to catch a recipient's attention and build more of that personal connection.

What we learned

Our schedules consist of taking CS classes after CS classes; however, this hackathon was the biggest learning experience for us. We learned a ton about ranking and similarity algorithms, how to prompt and fine-tune Gen AI models, and how to present everything in a full-stack application with working authentication and a backend database. On top of just the technical skills, we learned how to think from a user’s perspective and help drive feature development.

What's next for Luup

With Summer 2024 coming up, there’s only going to be one thought in every college student’s mind—Summer 2025 Internships! We’re trying to have as many students using Luup this cycle as possible!

We’ve got tons of feature ideas:

A passive search feature that automatically finds people for you to connect with in the background based on your interests and then displays them for you every morning.
A connection share feature that allows you to help other users who want the internship that you have, and in return for helping that user, you are then connected with someone with the internship that you want: promoting a supportive cycle of mutual assistance and opportunity sharing.
LinkedIn authentication to make onboarding easier
A customized message feature that would allow users to upload their writing to make the message sound more like themselves
The ability to have our Gemini model learn from the messages it creates by asking users how well the email preformed, and tons more

We’re passionate about solving this pain of networking in college and are excited to keep adding to what we have!