AI decision-making platform
Clarity / North AI
An AI assistant that answers from your own decisions.
A decision-making platform. People log decisions, run guided Clarity Break sessions and complete a profile assessment; North, the built-in AI assistant, answers from that history.
- My role
- Full stack engineering
- Product
- Web app · AI assistant
- Links
- Private client project
Overview
The product
Clarity helps people make and review decisions: a Decision Log, guided Clarity Break sessions and a profile assessment, with North, an AI assistant, built on top of all three.
Challenge
What had to change
General-purpose AI chat knows nothing about the person asking, so its advice stays generic.
Response
The system built
Every decision, session and profile result is embedded as it is saved, so North retrieves the user's own history before it replies and can point to patterns in it.
Product capabilities
What the product needed to do.
Log
A Decision Log for recording decisions and coming back to them later.
Reflect
Guided Clarity Break sessions for thinking a decision through.
Assess
A multi-module profile assessment with scoring, archetype results and module retakes.
Ask
North AI chat as a floating widget or a full page, with multiple conversations.
Subscribe
Paid access through Stripe subscriptions.
Technology
The stack behind the experience.
App
- Next.js
- TypeScript
- TanStack Query
- Redux Toolkit
Data
- Supabase
- PostgreSQL
- pgvector
AI and billing
- OpenRouter
- Stripe
My part
What I built.
Built the retrieval layer behind North, the multi-conversation chat, the multi-module profile assessment with scoring, and the Decision Log and Clarity Break flows on Next.js and Supabase.
Engineering notes
- Every decision, session and profile submission is summarised and embedded when it is written, then upserted into a pgvector knowledge base.
- At question time the query is embedded and matched by cosine similarity against an HNSW index, scoped to the signed-in user.
- Chat generation and embeddings go through one OpenRouter integration, so the model is a configuration value rather than a code change.
Delivered system
What shipped.
- Decision Log and Clarity Break flows
- Profile assessment with scoring and archetypes
- North AI chat grounded in each user's history
- Stripe subscriptions for paid access
More work