Zentrin

Grow your Financial Wisdom, fast-track your Financial Freedom.

Comment

Inspiration

To give people wisdom and tools to fast-track their financial freedom, so they can redirect their time, talent, and capital toward greater purposes that move society forward.

What it does

It provides:

  • A personal finance mentor, available 24/7, at a tiny fraction of the cost of a real mentor.
  • Goal and action plan tracking to pull the 4 levers towards financial freedom: the Wisdom lever to stay on track, and the Spend Less, Earn More, and Invest Better levers to fast-track freedom.
  • Key financial wisdom/freedom metrics: Freedom Age, Wisdom Score, Savings Rate, and Progress Score.
  • A financial situation dashboard (cash flow, assets, debts, risks, and shields).
  • A Financial Literacy learning syllabus and learning sessions applied to your own situation and numbers.

There are so many financial literacy resources on the internet today, which is fantastic, but the downside is that it's becoming difficult to separate the fundamental unbiased stuff (the signal) from the rest (the noise). And then to apply that knowledge to your own situation. And to do all that efficiently and effectively so you spend as little time on finance as possible and focus your life on what matters: your loved ones and your purpose. Zentrin seeks to solve these problems: The Mentor focuses on the fundamentals and proven frameworks, and then helps you apply them to your real actual financial situation/numbers. Like a GPS, the Zentrin App shows you the Map to get to financial freedom, and assists you on your journey to get there safely and quickly.

How we built it

Progressive Web App using Lovable as the AI coding agent, using the Gemini API for the Mentor.

Challenges we ran into

  1. LLM cost vs latency vs output quality trilemma: Had to iterate a lot with the LLM API calls architecture to find the sweet spot of acceptable mentor message quality, <5 seconds latency, and <1 cent per message (system prompt >20k tokens). Gemini 3 Flash (low thinking mode) was the only model config that satisfied the cost and quality constraints, but latency was extremely variable and totally unacceptable at times (>60 sec), probably because it was a preview model with low priority. This was a roadblock until Gemini 3.5 Flash Lite shipped in late July, but it arrived too late for this contest (to put the app in production and acquire real customers/revenue).
  2. Taming the Gemini LLM to behave like a mentor: No Financial Product Advice as defined by Australian law—the strictest in the world. Finance is highly regulated, including 'Financial Advice', especially in Australia. AI models have a tendency to give advice, so had to iterate a lot to 'tame' the LLM into mentoring behaviour (being neutral and NEVER influencing the user in their financial decisions). Objective-driven time-boxed mentoring sessions, where the Mentor drives the agenda and nudges to end the session, which is not natural for an LLM.
  3. Legal liability risk: Risk of being sued for causing financial harm. Had to register a limited liability company to shield myself and my family from that risk, which added significant administrative burden and delay.
  4. Payment processors: After having been through wiring Paddle as payment processor (Merchant of Record), testing it, and getting approved, they later came back and reversed their approval on the basis that the App had a chatbot and was related to finance. So had to find another MoR that would accept my app and go through the whole process again, which caused a delay of 2 weeks.
  5. Data security and privacy: Zentrin app captures people's financial numbers, which is sensitive data that needs to be secured and kept totally private. Had to spend a lot of effort to address that challenge.

Accomplishments that we're proud of

  1. AI Prompting: 80% of the work was spent iterating to build a production-grade system prompt (22k tokens) to overcome the first 2 challenges above, and orchestrating the LLM API calls into the sweet spot of the quality-latency-cost trilemma.
  2. App UI: Goals/Action Plans, financial situation picture, key metrics, learning syllabus, and badges.
  3. Business Registration: Created a real registered business entity and all the legal disclosures for the app (terms, privacy policy, etc.), user consents, etc.

What we learned

  • How LLMs really work (at the API level), the various ways to orchestrate/architect the API calls, and the limitations.
  • The Law (Financial Product Advice regulation, Privacy regulation, and the risks).
  • How to register/create a business entity.
  • The intricacies of payments (at a global scale).
  • How amazing vibe coding is!
  • And a ton more.

What's next for Zentrin

Advertise it, use real data to improve it (and use AI to automate improvements and run operations), then execute the long product roadmap we have in mind.

Built With

Share this project:

Updates