Angela from Accounting

Telling you no when you need to hear it

Comment

We live in a society...

Getting older means having more responsibility, and like it or not, handling money is a crucial component in that. With the recent release of Buut, a bank intended to guide 10-16 year olds to better handle their money, the average person's lack of financial knowledge is being emphasized. To help assist with this crisis, we created Angela from Accounting to help give Evas advice on how to best spend their money. Not only that, but to do so in a casual way that lands with young adults.

Angela from Accounting: Your financially savvy friend

The Problem

Most people in their 20s aren't bad with money because they don't care. They're bad with money because the moment they're standing in a shop, hand on a price tag is the one moment where zero financial tools actually help them. Budgeting apps live in the past. They tell you what you already spent. Bank apps show you a number. Neither one helps you in the aisle at H&M when you're holding a €80 jacket and genuinely unsure if you can afford it without draining your groceries budget.

In these moments, Angela is a friend you can rely on.

The Solution

Angela is an AI-powered budget advisor that lives inside a chat interface connected to real bunq accounts. She's not a dashboard or a spreadsheet. She's just like another contact in your phone you can confide in. She knows how much money you have, how you've been spending it, and isn't afraid to tell you the truth.

The core flow:

  1. Snap a photo of the item you're thinking about buying
  2. Angela identifies the product, price, and category, using multimodal vision AI
  3. She checks your real bunq balance for that category and gives you honest advice
  4. If the money isn't there, she tells you which budget has the slack and offers to move it All in one chat.

How It Works (Technical)

Multimodal AI pipeline (AWS Bedrock):

  • A vision model parses the product photo and returns structured JSON: item name, brand, price, and budget category matched against the user's live bunq accounts
  • A chat model (Claude via Bedrock Converse API with streaming) handles the conversation, with full tool-calling support for live banking operations

Real bunq integration:

  • Full 4-step authentication flow: sandbox user creation → installation → device registration → session
  • Every API request is signed with RSA-SHA256 over the raw request body — proper cryptographic auth, not mocked
  • Each budget category is a real bunq savings account, not a label. Transfers between them are real API calls in the bunq landscape
  • Transaction history is fetched live and filtered to exclude internal transfers, so Angela's spending analysis reflects only real external expenditure

Chat architecture:

  • Streaming responses via Server-Sent Events, meaning Angela types in real time
  • Responses split into short rapid-fire bubbles using a --- separator, simulating a text conversation with a real friend
  • Tool-calling loop with error isolation, so a failed tool call never corrupts the conversation history
  • Session memory persists across messages within a conversation

User-defined budget tiers:

  • On first open, the user assigns each of their bunq categories to one of three tiers: Protected (savings, never touch), Essential (groceries, health), or Leisure (clothes, entertainment)
  • Angela uses these explicit assignments in every recommendation, so she knows not to suggest pulling from Groceries to fund Entertainment, because the user told her Groceries is Tier 2

Why Angela Stands Out

Most financial tools treat money as data. Angela treats money as a conversation. A few things that separate this from the obvious approach:

  • The photo-first paradigm. Nobody opens a budgeting app in a shop. But everyone opens their camera. - Snapping a photo is frictionless. It's the same gesture as sending a friend a picture of something you want. Angela makes the next step just as natural.
  • Personality that actually lands. Angela doesn't talk like a bank. She texts like a friend who happens to be good at accounting. Dry, warm, brief, honest. The target audience (18–25) will close an app the moment it starts sounding corporate, so Angela isn't that lol.
  • Rebalancing, not just advising. Most budget advisors stop at "you can't afford this." Angela goes further: she finds the slack in your Tier 3 accounts, suggests a specific transfer with a specific amount, and executes it in the same conversation. The advice is actionable in seconds.
  • Tiers built by the user, not guessed by the algorithm. Angela doesn't assume groceries matters more than clothes, the user tells her. This makes recommendations feel personal rather than prescriptive.

Biggest Strengths to Highlight

  1. Decision-point UX — Angela helps at the moment of purchase, not after the fact. This is the gap no other tool fills.
  2. Real bunq integration — Actual account creation, cryptographic request signing, live balance reads, real transfers. Not a prototype.
  3. Multi-modal — Camera, voice, and text all work. The photo flow is the star, but the system is genuinely multimodal.
  4. Character matters — For 18–25s, personality is a feature. Angela's voice is a product decision, not a garnish.
  5. Tiers = personalisation without complexity — One quick onboarding gesture gives the AI enough context to give advice that feels specific to you, not generic.
Share this project:

Updates