Kinlia AI Event Discovery

• 

  Kinlia Architecture

Inspiration

Inspired by Burning Man and the amazing community and connections, we want to help address the huge decline in people gathering in person. Today people have fewer friends. They spend many more hours at home. They host friends less and go to social events much more infrequently than 10-20 years ago. This decline in gathering in person negatively affects people physical and emotional health.

What it does

Kinlia AI scraps thousands of in-person events. In learns a person interests based on history and chats with AI, and it recommends events based on that and based on what events their friends are going to . The AI acts like an uber social host, connecting people in groups (and nudging them) to go to events together in two products : Crew Up which invites friends and strangers to attend events together, and Collab, helping you coordinate to attend events with your friends.

How I built it

It is built in next js in the front end, phython in the back end. WE are using Apify with N8N to scrape events from approved hosts (to insure event quality). We are using Postgres + pgvector to store event and user embeddings for fast similarity search. Python (FastAPI) to runs the AI recommender backend. Sentence-Transformers (E5-small) to creates embeddings from event text and user interests. BGE Reranker (Cross-Encoder) to re-order top matches for higher precision. Llama 3 8B (via Ollama) for small local LLM for vibe tagging and AI explanations., and TypeScript API to connects the mobile app & web frontend to the recommender

Challenges I ran into

Bringing evertying together. THere are lots of moving pieces.

Accomplishments that I'm proud of

Scrapping events with APify to publish on website

What I learned

Persistence is key.

What's next for Kinlia AI Event Discovery

Enabling our LLM to chat with multiple people at once as it nudges them to go to events together!

Built With

• aipify
• n8n
• nextjs
• phython
• vs
• warp