Northstar is a co-built financial-planning prototype that turns a synthetic portfolio and goals into deterministic stress scenarios, visible memory traces, and plans that require explicit approval before execution.
Stack, status, evidence, and public actions are rendered from the typed project record.
Category
AI Systems
Type
App
Priority
Strong
Overview
What this project is
Northstar is a co-built financial-planning prototype that turns a synthetic portfolio and goals into deterministic stress scenarios, visible memory traces, and plans that require explicit approval before execution.
Problem
Why it matters
Financial-planning agents are difficult to trust when their assumptions, memory, calculations, and action boundaries are hidden behind a chat response.
Solution
Approach
A memory-first workspace that separates deterministic scenario math from agent narration, exposes the evidence behind each plan, and keeps consequential actions behind a human approval gate.
Architecture
System shape and stack
Methodology
Deterministic before persuasive
The checked-in synthetic fixture uses a fixed seed and as-of date. Scenario outputs come from documented asset-class shocks and cash-flow arithmetic rather than generated numbers.
Trust Boundary
Identity and ownership stay server-enforced
The API verifies bearer tokens, rejects cross-user identities, forwards the user JWT to Supabase, and includes a migration from permissive prototype policies to owner-scoped row-level security.
Agent Design
Reasoning is visible; execution is gated
Plans carry an inspectable memory and scenario trace, while state-changing execution remains pending until the user explicitly approves it.
Contribution
A team build with commit-backed ownership
Northstar was co-built with Kushagra Bharti. Yuvraj's visible Git history covers much of the route and product integration, authentication presentation, onboarding gates, user-specific memory and agent surfaces, scenario behavior, goal mutations, and build recovery.
- React
- TypeScript
- Express
- Supabase
- Python
- OpenAI Agents SDK
Technical Highlights
Visible technical signal
- Deterministic portfolio stress scenarios with checked-in provenance
- Bearer-token verification and owner-scoped Supabase access boundaries
- Inspectable reasoning trace with explicit approval before execution
What It Proves
Builder signal
Ability to integrate frontend, API, data-security, agent, and reproducible-analysis concerns into one inspectable product system.
Boundaries
Context that should stay visible
No live demo is claimed. The checked-in scenario uses synthetic data and is not a forecast, tax calculation, trade recommendation, or account connection. The historical Supabase project is no longer reachable, so its owner-scoped RLS migration is documented and tested in code but not verified against that remote project.