NextSpot: Review-Powered Recommendations

See a live demo of NextSpot, which uses LLMs to understand review signals for precise, category-constrained local recommendations without keyword searching.

Overview

NextSpot is a live web demo that recommends similar places in a new city by understanding what your favorite spot is known for from public signals (reviews + description).

In the demo, I’ll enter a favorite place and a target location (e.g., Dallas/Houston). NextSpot extracts the place’s “signature attributes,” generates a compact intent query, searches locally within a strict radius, and returns recommendations constrained to the same business category (e.g., cafe → cafe, salon → salon). The result is a recommender that works across many categories without hardcoding for any single domain.

In 5 minutes, I’ll show the working web app and a short walkthrough of the backend internals:
1) pulling place metadata + top reviews,
2) extracting “signature attributes” into structured fields (not freeform text)
3) turning those fields into a constrained search query,
4) retrieving candidates from Places within a hard geo-radius, then
5) ranking results with embeddings + a lightweight re-ranker

Tech stack