Similitude

Overview

Most platforms reduce an experience to a single number: a star rating, a thumbs up, a score. Similitude captures the whole shape of a reaction instead. Users react in real time with a continuous slider while something unfolds (a concert, a debate, a city council meeting, a video, a meal), and the platform records the complete time-series curve of their response. A proprietary similarity engine then compares those curves across users to surface people who felt the highs and lows at the same moments.

The result is a social network built on behavioral kinship rather than demographics or follower counts. Beyond live experiences, Similitude includes a posts-and-feed system where every post carries the author’s slider position, threaded comments, real-time private messaging, gamification, and location-aware discovery of more than 158,000 seeded subjects across the Seattle/Tacoma and Reno/Lake Tahoe regions.

My Role

Founder, sole product owner, and project director. I conceived the product, defined its privacy model and similarity methodology, and directed every phase of design and development across twelve major build sessions. Similitude was built using an AI-assisted development workflow: I directed system architecture and specification work with Claude, then translated each feature into tightly scoped implementation prompts executed by Claude Code. I owned all product decisions, approval gates for production changes and deployments, infrastructure troubleshooting, and the full documentation set (product spec, technical architecture blueprint, database schema blueprint, and session handoff notes).

Key Features and Engineering Highlights

The similarity engine (now at v1.6) compares wall-clock-anchored and media-position-anchored reaction curves using RMSE on a 5-second interpolated grid, blended with topic behavior, post and comment agreement, and a private importance signal. Raw scores are never exposed; users see only qualitative labels.

Privacy is enforced by design. Users are pseudonymous, individual reaction timelines and values are never visible to others, importance signals and personal notes are always private, and aggregate data appears only after anonymity thresholds are met. Analytics and CAPTCHA tools were chosen specifically for their privacy posture.

Recorded media support embeds YouTube, SoundCloud, and HTML5 audio/video, tying every reaction point to its exact position in the media so curves can be compared across viewers who watched at different times.

Offline-tolerant reaction capture uses IndexedDB with exponential-backoff syncing, so no reaction data is lost on a flaky connection.

Real-time messaging runs on self-hosted WebSockets, and a graduated, dynamically computed post-visibility model adjusts automatically as the network grows.

Technical Stack

Backend: Laravel 10 (PHP 8.3), MariaDB, Redis (cache and queues), Laravel Sanctum, Laravel Horizon, Laravel Reverb (WebSockets), Laravel Cashier (Stripe)
Frontend: React 18, TypeScript (strict mode), TailwindCSS, React Router, TanStack Query, Zustand, Vite, Headless UI, Dexie (IndexedDB), Leaflet, react-joyride
Data and services: Overture Maps Foundation data (via DuckDB), Mapbox geocoding, Anthropic Claude API for automated topic inference, Elastic Email
Infrastructure and monitoring: Cloudways (DigitalOcean), Cloudflare CDN and Turnstile, Sentry, Plausible Analytics
Development approach: AI-assisted architecture and implementation with Claude and Claude Code

Status

Fully built and live, currently in its user-acquisition phase, with ongoing feature development.