Skip to content
Anastasiia Shulga.
  • Work
  • About
  • Contact
design@anastasiiashulga.comLinkedIn ↗© 2026 Anastasiia Shulga
All Work
AI ProductConceptTry demo

Lodestone

A conversational finance assistant where the answers are generative UI cards, not paragraphs. Trust shown, not claimed.

Role

Product, Interaction, System

Year

2026

Scope

Conversational UI · Cards · Trust patterns

lodestone

Direction · Sources · Confidence

Full product

The interactive experience

Portfolio overview, conversation thread, and generative card response — all in one view. Every number sourced. Every claim traceable.

app.lodestone.finance · portfolio view

Product

A live conversation

Step through the demo. Each prompt builds the next answer in place with its own card primitive, source pills, and confidence sliver.

Lodestone Workbench · click a prompt to advance the thread

Pattern library

Card primitives are composed per question.

Allocation

Compare

Cashflow

Trust UI

Three signals that earn the user's trust

The app

Lodestone in your pocket

Same brain, thumb-sized. The center phone plays a typical minute on loop — asking in plain language, nudging a what-if, and correcting a number when life changes. The answers recompose; the sources stay attached.

Try it yourself

Ask Lodestone a question

Type your own question or pick a prompt. Lodestone reads it, pulls from the linked accounts, and composes the card that fits the answer — sources and confidence included. Try asking whether to buy something and watch it decline: data, not advice, by design.

Live demo · Lodestone composes the card, sources, and confidence per question

Process

Three decisions that shaped Lodestone

The demos above look inevitable in hindsight — they weren't. These are the calls that defined the product, and the options they had to beat.

Decision 01

Cards, not paragraphs

Considered

A plain chat — money answers arriving as well-written text, like every other assistant.

Chose

Composable card primitives (allocation, compare, cash flow) that the model selects and fills per question.

Because

Numbers in prose are unverifiable. A card carries its chart, its sources, and its confidence with it — trust needs a surface to live on.

Decision 02

Trust is a UI layer

Considered

Citing sources on request — a tooltip, an expandable footnote, an “explain this” button.

Chose

Source pills and a confidence sliver on every single answer, unhideable.

Because

Trust claimed in copy is worthless; trust shown per-answer is the product. If the assistant read your accounts, you should see exactly which ones, every time.

Decision 03

It says no

Considered

Letting the assistant answer everything, hedged with disclaimers in fine print.

Chose

A hard advice boundary — “data, not advice” — plus honest low confidence on thin data.

Because

A finance assistant that answers “should I buy X” is a liability, and users can feel it. Refusing precisely is the credibility move.

← PreviousTerrainAI ProductNext →ApexAI Product