Inspiration

Wealthy families have a financial adviser, a plan and regular money conversations around the kitchen table. Most families have none of these. Knowledge about pensions, gifting, ISAs and scams rarely passes between generations, so young people learn about money from social media while older relatives are the main targets of scams. We wanted to give every family the kind of planning conversation that usually only money can buy.

What it does

Heirloom is an AI money coach for a whole family across three generations: a grandparent, a parent and a young adult.

  • Next 90 Days forecasts each person's balance with an uncertainty band, their lowest point, and their chance of going below £0.
  • Early warning shows why someone is at risk, factor by factor, and offers free, independent help (MoneyHelper, StepChange, Citizens Advice) before anything else when the risk is high.
  • Family Time Machine runs 1,000 simulated futures over 25 years to compare giving money now, in 10 years, or as an inheritance: how likely the young adult is to reach a house deposit, and how likely the grandparent's savings are to stay above a safety floor.
  • Voice Coach answers money questions out loud in the family's own language, using the family's real forecast rather than generic advice.
  • Claim Check rates money tips from TikTok, Instagram or group chats, flags scam warning signs and links to sources.

Our second example family, the Okafors, has nothing to inherit, and Heirloom still helps them: small regular saving, the Lifetime ISA bonus and avoiding debt traps. Heirloom gives guidance and scenarios, never regulated financial advice.

How we built it

  • Frontend: a single-page app with a magnifying dock for navigation, one page per tool, and a day/night theme. Each family member has one colour everywhere they appear, so judges and users can follow a person across every chart.
  • Forecasting: Monte Carlo simulation of daily income and bills (400 paths per person) for the 90-day forecast, and 1,000 simulated futures with assumed returns and volatility for the Time Machine. [Add if built: LightGBM quantile model and SHAP explanations.]
  • Backend: [FastAPI with /forecast, /timemachine, /claimcheck and /ask endpoints; the frontend falls back to in-browser simulation if the API is unreachable.]
  • Voice: [ElevenLabs agent that calls our forecast API as a tool] / browser speech in the prototype.
  • Claim Check: [Firecrawl snapshots of gov.uk and MoneyHelper pages, plus the FCA Register API] / scam-pattern rules in the prototype.
  • Data: synthetic transactions for two families, with assumptions informed by ONS earnings, life-expectancy and household-spending data.

Challenges we ran into

  • Making uncertainty understandable. "A 14% chance of dropping below £0" and shaded likely-range bands worked far better than a single predicted balance.
  • Staying on the right side of financial regulation. Every answer is framed as a scenario, and anything high-risk is handed to free human help rather than a product.
  • Keeping the numbers honest. We kept all the maths in deterministic code and only let AI explain the results, because language models can produce confident wrong numbers.
  • Repeatable demos. The forecast depends on today's date, so the numbers shifted from day to day. We had to pin a date for the demo.
  • Design. Our first version was a long scrolling dashboard of boxes. We rebuilt it into separate pages with a clear colour system, which made it far easier to present.

Accomplishments that we're proud of

  • A Time Machine that turns a hard family conversation ("should Gran give Zofia money now or leave it in the will?") into two clear numbers.
  • Showing that Heirloom helps a family with nothing to inherit, not just wealthy ones.
  • Explanations behind every warning, so nobody is told they're "high risk" without knowing why.
  • A demo that still works offline.

What we learned

  • How Monte Carlo simulation turns uncertain futures into probabilities people can act on.
  • The difference between financial guidance and regulated advice, and how to design around it.
  • That good design is mostly about removing things: fewer boxes, one colour per meaning.
  • Practical teamwork: API contracts agreed in the first 30 minutes, feature branches and pull requests.

What's next for Heirloom

  • Connecting real accounts through the NatWest Bank of APIs with Open Banking consent for each family member.
  • [Replacing the simulated forecast with trained models and SHAP explanations, plus fairness checks across income bands and regions.]
  • Adding more languages to the voice coach, starting with those most spoken in UK homes.
  • A "family meeting" mode that prepares an agenda and talking points for the whole family.
  • Early-warning signals that let bank support teams reach out before someone falls into hardship.

Built With

Share this project:

Updates

Submission history