I solo built the entire automation engine behind Wealtii, a crypto index fund platform. Here is how it actually works.
When I started building Wealtii, I wanted to solve one problem: why does diversified investing in crypto and tokenized assets still require so many manual steps?
If you want exposure to Bitcoin, tokenized gold, and US tech stocks today, you need accounts on multiple platforms, manual purchases of each asset, your own spreadsheet to track weights, and the discipline to keep doing it. Most people give up after step two.
So I built an automated system that handles the entire flow.
Here is what happens when a user invests $100 into one of our funds:
1. They deposit USDT on BNB Chain to a unique address generated for their account
2. A custom blockchain watcher I wrote detects the deposit in real time
3. The backend calculates the exact amount of each underlying asset to purchase based on the fund's target weights
4. It executes all the swaps automatically across the relevant on-chain venues
5. The purchased assets are settled into a public Gnosis Safe multi-sig vault
6. The user receives fund units and can verify the 1:1 backing on-chain at any time
That entire flow runs without a single manual step from me or the user.
One click. Seven assets purchased. Everything verified on-chain.
The three live funds hold a mix of crypto (BTC, ETH, BNB), tokenized gold (Tether XAUT), tokenized US equities (Nvidia, Apple, QQQ, S&P 500 via Ondo Finance), and tokenized silver.
Building this solo meant writing:
- A FastAPI backend with Celery task queues across 6 dedicated queues
- A custom blockchain deposit watcher
- Web3(.)py integration with the Gnosis Safe SDK for multi-sig custody
- A React frontend with real-time WebSocket portfolio tracking
- 9 Docker containers orchestrated with Docker Compose
- Full Terraform IaC for AWS deployment
- CI/CD pipelines for automated testing and deployment
336,000 lines of production code. No co-founder.
The next major milestone on our roadmap is automated periodic rebalancing, so the funds will not just execute the initial allocation but also maintain target weights over time as markets move.
After that, AI-assisted allocation adjustments.
I am building these layer by layer because each one touches real user money and needs to work perfectly before it goes live.
If you are building anything in fintech or web3, I would love to hear: what was the hardest automation challenge you faced where "it works on my machine" was not good enough because real money was involved?
Currently live at wealtii.com with 0% platform fees.
Agiloop's point about evidence before automating matches where we ended up with an AI agent in one of our own apps. Ours doesn't move money, but some of its actions can't be undone either.
Users can set most agent actions to "always allow". Irreversible ones never get that option: the agent proposes, a person confirms every time, and the app enforces it rather than the prompt. We also log for every agent write whether it needed no approval, ran on an always rule or was confirmed by a person, and which model made the call. We added that late, after noticing the log could say what happened but not why it was allowed.
For the AI-assisted allocation: will it propose changes someone confirms, or trade on its own within limits?
For systems moving real money, “it works” isn’t enough. I’d want independent evidence around security, failure recovery, compliance and operational readiness before automating rebalancing. That’s the kind of production-readiness gap we built Agiloop to assess.
With three funds live, what investor behavior or portfolio drift would justify prioritizing automated rebalancing before AI-assisted allocation?