Product · Engineering · Responsible AI
Staying on top of my financial life had started to feel like another job.
I'm comfortable with money and with technology. What I didn't have was time: time to check five apps, keep a spreadsheet going, and piece the full picture together every month. So I started building something that would do that work for me.
01 · Why Valenso
I needed less financial admin.
It turned out I wasn't the only one. When I talked to friends and family, I heard the same thing again and again. People who understand their money still spend hours checking apps and rebuilding the same picture by hand.
Tools like Monarch and Rocket Money already put your finances in one place. I wanted Valenso to go a step further: do the organising for you, understand where all the money actually moves, and flag what needs a look, while always showing its working and leaving the decisions to you.
- Built for
Busy people who care about their money.
Several accounts, a few credit cards, money moving in every direction. You know what matters. You just don't want to spend your weekends keeping the system up to date.
- AI-first, not AI-added
AI sorts the transactions, learns how you think about money, and notices patterns worth knowing about, so the app comes to you instead of the other way round.
- Still in control
Money and black boxes don't mix. Valenso tells you what it did and why, learns when you correct it, and leaves the final say with you.
Why now
AI agents finally made it possible to build.
I'd wanted to fix this for years. What changed was AI agents that can turn a carefully written plan into working software, fast enough to build the whole system, not just a demo.
It was a chance to put my product, engineering and AI experience to work together. The agents write much of the code. I make the product and architecture decisions. And I built it to solve my own problem first, on foundations secure enough to open up to others later.
- Product judgment
Decide what to build, and why, before any AI writes a line.
- Engineering discipline
Keep architecture, security and releases in human hands.
02 · The product
A clearer, more confident way to stay on top of your money.
Valenso connects to your bank accounts and credit cards, or you can upload statements. It sorts every transaction into one story you can follow: what came in, where it went, what you saved, and how much is left to spend.
It's honest about what it knows, and clear about what it doesn't. Every automatic category comes with a reason. Your corrections stick. Possible duplicates are pointed out, not quietly merged. And every number can be traced back to the transactions behind it.
The guiding principle
A simple test for every decision.
Kanso is a Japanese design idea: remove what isn't needed, so what matters stands out. I use it to judge everything I add to Valenso.
Does this remove financial clutter and make what matters easier to see?
- Design
- Calm and uncluttered.
- Language
- Plain words, without dumbing down.
- Data
- Numbers you can check.
- AI
- Help, not autopilot.
From why to how
AI made it fast. Experience made it sound.
I've spent years building financial products at a Fortune 500 company, in a regulated industry. That shaped how I built Valenso. AI agents let me move fast across product, design and code. The parts that make it trustworthy came from that experience: clear customer journeys, accurate money logic, firm security boundaries, and someone accountable for every decision.
- Product
Start with the real problem
Start from a real frustration, and turn it into clear goals.
- Engineering
Make the numbers dependable
Keep the money logic exact, traceable and protected.
- Agentic engineering
Use AI to expand the build
Brief the agents well, check their work, and use them everywhere.
- Guardrails
Set non-negotiables early
Settle what data can move, and what AI can see, before building anything.
03 · The build
Fast, but not improvised.
How I built Valenso mattered as much as what I built. AI set the pace. A simple routine kept it honest.
Write it down first
Before an agent builds anything, I write down the problem, what it should do, the lines it must not cross, and how we'll know it works.
Build in small, checkable pieces
Agents turn each decision into working code, tests and notes. I review every piece, as both product owner and engineer, before it goes further.
Use it, then re-plan
Using Valenso every day shows me what to fix next better than any roadmap. When the evidence changes, so do the priorities.
AI shortened the path from idea to working software. It didn't replace judgment, or the care that goes with being responsible for it.
04 · Architecture
Your numbers stay right, even when the technology changes.
Valenso keeps its money logic separate from the companies that supply bank data and AI. That means we can swap a provider without touching your numbers, test every rule, and draw a hard line around your sensitive information.
- EXPERIENCE – What the user sees – Dashboard · transactions · guided workflows
- DOMAIN CORE – Financial logic – Money · reporting · matching · user choices
- DECISION SYSTEM – Explainable rules – Ordered matching · clear reasons · reversible choices
- PRIVATE DATA – Encrypted financial data – One user at a time · protected fields · audit history
- BANK CONNECTIONS – Replaceable providers – Secure sync · common format · duplicate handling
- CONTROLLED AI PATH – One route to AI – Approved fields · redaction · cost limits · activity log
Change providers without a rewrite
Bank-data and AI providers connect at the edge; Valenso’s financial logic stays independent.
Keep AI out of financial truth
Your money, matching and reports are handled by tested code. AI can make suggestions. It never decides your numbers.
Make releases reversible
Small database changes, feature switches and separate environments make it safer to learn and correct course.
05 · Security & governance
Trust was designed in from day one.
With financial data, the stakes are high. So Valenso treats outside AI companies as untrusted. It decides exactly what may leave, records why, and keeps anything sensitive inside.
The golden rule
No personal or sensitive financial data goes to the model.
Nothing leaves by default. Something can be sent only once it's been explicitly approved and checked for personal or payment details.
Send the minimum
Models see only what they need: a cleaned-up merchant name, never your amounts, balances, account numbers or personal details.
Use one controlled route
Every AI request follows the same guarded path. Tests prevent a new feature from quietly bypassing it.
Record what leaves
Valenso records the purpose and shape of every AI request without logging the sensitive value itself.
Control cost
AI usage is measured and capped; bank-provider usage is recorded so the cost of running the product stays visible.
Keep people in control
AI answers are suggestions. The user confirms the durable category, tag or rule, and can always see why it happened.
The wider security posture
Encrypted sensitive fields, strict per-user queries, MFA-gated account linking, signed provider webhooks, content security policy, secret scanning and isolated environments reinforce the AI boundary.
This describes how Valenso is built. It isn't a security certification. The controls are written down, tested, and will keep evolving.
Continuing the build
Valenso is still a work in progress, and this is the story so far.
I keep improving it by using it, and I write down the decisions and lessons as I go. If you're curious about any part of it, I'm always happy to compare notes.
Next, I'm working on asking Valenso questions in plain language, like "can I afford a trip to Japan this year?", answered from your own plan, without your numbers ever leaving Valenso.