2026 · Product Lead & Designer

Whspr

Gender-aware urban safety intelligence, powered by community

The Problem

Urban safety apps are generic — they don't account for time of day, your gender, or the lived experience of the people who actually use a space. Whspr crowdsources hyperlocal safety signals, layers them with real NYPD crime data, and generates AI summaries tailored to day/night context and gender identity.

Role

Product lead and designer — tech stack selection, database schema, anti-bot trust scoring system, AI integration strategy, and MVP scoping.

Timeline

6 sprints

Tools & Methods

  • Next.js + Vercel
  • Supabase (Postgres)
  • Mapbox API
  • OpenAI GPT-4o-mini
  • NYC Open Data (NYPD)
  • Figma

The Product

Users search any NYC venue or neighborhood via Mapbox autocomplete. The place profile surfaces crowdsourced signals, 90-day NYPD crime statistics, and a live AI-generated summary that regenerates when you toggle day/night or gender filter. The contribute flow collects signals through a 3-step form after a lightweight sign-up that captures gender once, not on every submission.

Trust Scoring System

Whspr's anti-bot layer scores every signal in real time: −0.2 for accounts under 7 days old, −0.3 for burst submissions (5+ signals per hour), −0.5 for 2+ community flags. Signals below 0.4 are hidden automatically — not deleted, kept for audit. This keeps the feed credible without requiring moderation.

Key Design Decisions

  • Gender field on the users table (not signals) — consistent identity, single capture

  • Real NYPD crime data over synthetic alternatives — academic credibility matters for a thesis

  • GPT-4o-mini for LLM summaries — cheapest, fastest model that still handles nuanced tone

  • NYC Mapbox bounding box to prevent scope creep in the MVP

  • Seed data: 5–6 venues with 4–6 realistic signals each — quality over quantity

Four-Table Schema

  • users — auth, gender, account age for trust scoring

  • places — Mapbox POI data, geocoordinates, neighborhood

  • signals — crowdsourced submissions with trust score and flag count

  • routes — optional route-level safety for v2 expansion

Outcomes

3Production screens
2Real data sources
3Trust score variables
6Sprint delivery