Quotient

Overview

Quotient is an interactive self-insight tool: you take an adaptive quiz across eight cognitive domains, get an animated results reveal, read a plain-language report, and can ask a chatbot questions about your own results. It is explicitly an entertainment and self-reflection project — not a clinical or diagnostic instrument — and the app says so in the UI, the report and the chatbot's system prompt.

It was built over two days as a sequence of small phase-by-phase pull requests, which made it a good exercise in keeping a feature-rich app understandable.

What It Does

• Eight domains: logical, numerical, verbal, spatial, memory, processing speed, working memory, attention
• Adaptive difficulty: a transparent staircase that moves difficulty up after a streak of correct answers and down after misses
• Scoring: per-domain raw score → z-score against a cohort bucket → 100/15 display scale → percentile, plus a composite
• Report: rule-based plain-language report, with an optional LLMLLMLarge language model — a neural network trained on large amounts of text to predict the next token, used for generation, summarisation, classification and more. pass that restyles the prose but cannot change numbers
• Chatbot: retrieval-augmented, grounded in a local knowledge base and the user's own results
• Gamification: badges, streaks, confetti, between-domain interstitials, four selectable themes plus a glass theme
• Dashboard: history, progress charts and a radar chart of domains
• PDF export of the report, and a PWAPWAProgressive Web App — a web app that can be installed and keep working offline, using a web app manifest and a service worker. shell with offline fallback
• Accessibility: skip-to-content link, keyboard navigation, prefers-reduced-motion support and AA-contrast theme tokens

How It Works

1. Session engine: each answer is stored with the difficulty it was served at; the next difficulty is computed from that domain's history.
2. Scoring: the raw score is the mean effective difficulty of a domain's responses, because an adaptive test pushes everyone toward similar accuracy — what differs is the level they settle at.
3. Report: a deterministic generator builds the report; an optional OpenRouter pass rewrites only the narrative fields.
4. Chat: the question is embedded locally with all-MiniLM-L6-v2, searched in an embedded LanceDB store, and the top snippets plus the user's results go into the system prompt.

Tech Stack

Frontend: Next.js 16 (App RouterApp RouterNext.js's routing system built on the app/ directory and React Server Components, with nested layouts and data fetching on the server.), React, TypeScript, Tailwind CSS, Framer Motion, Recharts, Zustand

Data / AI: Prisma 7 with SQLite (driver adapter), LanceDB, @xenova/transformers (local embeddingsembeddingA list of numbers (a vector) representing a piece of text, an image or audio so that similar meanings land close together — the basis of semantic search.), OpenRouter for the report rewrite and chatbot

Export / Quality: puppeteer-core with serverlessserverlessA hosting model where the platform runs your code on demand, per request, and manages servers and scaling for you; instances are short-lived and keep no local state. Chromium for PDF; Vitest unit tests for scoring, adaptive logic, report generation and retrieval

Honest Notes

• The normative baselines are synthetic and illustrative, not derived from a real psychometric study, so percentiles are a demo of the maths rather than a measurement.
• The question bank is a seeded set of hand-written items (48 in the first pass, six per domain), so repeated attempts will see repeats.