Gossip - Reimagining Internal Company Communication

Team photo
DEV POV

Inspiration

We were driven by the idea of using embeddings to store meaning, unlocking powerful relevance matching. Our goal was to reinvent workplace communication by creating a system that connects people with the information they need, whenever they need it. We envisioned an experience where project updates, messages, profiles, and files could all coexist meaningfully within an embedding space. By using vectors as queries, we could surface relevant information, ensuring that projects or updates related to someone’s expertise stand out, while less relevant items stay out of sight. This vision offers visibility into a company’s knowledge and resources like no other service has before.

What it does

Our app organizes all workplace information—from messages and project updates to files and profiles—into a cohesive, queryable vector space. By embedding this data, our app enables instant recommendations based on relevance. For example, when someone searches or interacts with the platform, they see projects and updates most aligned with their expertise and interests, cutting through noise and bringing meaningful insights to the forefront.

How we built it

We leveraged embeddings and vector databases to create this unique platform. Using PostgreSQL with pgvector support, we built an underlying architecture that stores data as vectors, enabling quick and relevant search and retrieval. Our backend handles embeddings and recommendations, while our frontend organizes and delivers this information with a evolving user experience.

Challenges we ran into

Our biggest challenges came in setting up and developing the frontend to bring our vision to life. Although we quickly built a proof of concept, moving it to production revealed many obstacles. Developing an efficient, scalable system was complex, especially as we worked to master PostgreSQL and vector databases with no prior experience.

Accomplishments that we're proud of

We’re proud that we took our concept from theory to practice, successfully building a system that classifies similar items while keeping irrelevant data out of sight. Learning PostgreSQL and implementing it with vector search was a huge accomplishment for the team, especially given our limited experience. We might not have achieved perfection, but we realized our vision, which is what matters most. Considering this was our first hackathon, we are incredibly happy with the results.

What we learned

This project was an awesome learning journey. From PostgreSQL and pgvector to middleware implementations, we expanded our knowledge of how to store, query, retrieve, and display data in meaningful ways. We also deepened our understanding of practical, real-world problem-solving, gaining insights that will benefit future projects.