Interweave

Main page
Profile Page

Inspiration

I was inspired to build this project to address a very real problem that seems to be affecting a lot of people lately, but to be perfectly honest, I would simply love to use this product myself. No pun intended, but apparently I am not alone in feeling, at times, alone.

As someone who's spent countless hours behind a screen, I've seen firsthand how the current digital age has left many of us feeling more isolated than ever, myself included. The pandemic only made things worse, pushing those already teetering on the edge of introversion further into their own bubbles. I was never a party guy, but ever since March of 2020, I've been mostly at home, and whatever pressure there was to socialize in an office setting got removed by the ability to work remotely.

Fast forward to now, and it seems that my ability to make new connections has pretty much atrophied. Modern social media gave us algorithms to feed our specific interests but did little to nurture real human connections. There are interest groups online, but they generally revolve around a singular subject. There are dating sites, but they focus on rather short bios that cannot capture the truly complex pattern of a singular individual.

That’s when it hit me: A one-dimensional approach to connect won't cut it for multidimensional beings. We need something that understands and utilizes the depth of who we are.

You know what else is multidimensional? Vector databases.

It will all make sense in a bit, I promise.

What it does

Interweave takes the idea of connection over the internet to the next level. It doesn’t just find people who share a hobby with you, like a Facebook group might; it weaves together the full tapestry of the personality. In our individualistic culture with infinite freedom of choice, we become so wonderfully idiosyncratic, but available solutions like dating sites and meet-up groups are missing that aspect entirely, focusing on a very narrow view.

In simple words, the way Interweave works is like this:

You chat casually with a chatbot, which in turn pieces together a profile based on your unique blend of interests—be it Brazilian jiu-jitsu, an obsession with post-rock, or a fondness for vintage sci-fi. It identifies the level of interest and relevance for each subject, using natural language processing. Basically, it creates a model of yourself, fragmented into features that are stored as numerical values in a vector database.

The system then matches users whose profiles appear to be closer in the vector space, taking into account all of the values for their interests.

Technically speaking, it’s a semantic search for personalities.

The front end is minimalist—just a simple chat interface, a profile page where users can adjust their interest levels and relevance, and a connections page where they can message their matches.

Interweave is like a friendship finder, but smarter, deeper, and far more tailored. Not to speak of the more obvious application of finding a potential soulmate.

How we built it

I'm still working on the prototype as I'm typing this, but it has all been done here on the spot, talking to myself and a number of language models. The foundational idea of Interweave turned out to be much harder to implement than I expected, and perhaps the decision to go alone was a flawed one, but I regret nothing. It was a challenge; besides, it turned out to be a meta statement of sorts, considering the subject matter.

Challenges we ran into

The hardest part turned out to be making an LLM to act sane as a chatbot while using a smaller model in my local environment, while simultaneously making it deliver the required values to update user's profile in the database.

Accomplishments that we're proud of

-The pitch presentation is likely the best part of the project at this point. There is still some time left to complete the prototype, but the idea itself is what invigorates me the most. There is a lot of potential fun to be had in the future with Interweave, developing the algorithm to identify the most similar profiles as well as crafting features.

-Aside from helping the world of loneliness, it has great potential for earning by reimagining the approach to targeting ads—weaving these two concepts together. That would be the second accomplishment.