Summary
An LLM wrapper that gives opinionated, experimental investment advice. The opinions are the point; the experiment is whether opinionated framing makes the output more or less useful than a hedged one.
It is, to be clear, not financial advice. It is a small site that pretends to give financial advice — somewhere between a thought exercise and a toy.
The build is also an excuse to keep current on the practical edges of LLM application design: prompt structure, caching, latency, where to fall back to deterministic code, and how to make a model with opinions still sound like itself across a conversation.
Tech and approach
It started as a terminal tool. I exported a bunch of financial data from my bank accounts and fed it into a prompt that considered my multinational situation and a pile of specifics. It worked well enough for me, but when I wanted less technical people (my parents, basically) to try it and give their opinions, the terminal wasn't going to cut it. So I vibe-coded a frontend with a simpler prompt mechanism for less accurate but more accessible advice, aimed at people who want to test the idea without entering all their financial details.
I also used the project to learn about Electron, giving users the option to download a native app and keep all their data locally instead of uploading it to my database.
The stack is a React SPA with Tailwind, Postgres, and SuperTokens for auth. I wanted full autonomy: one monolith, no extra services or complications beyond the LLM runtime itself. That's what I got.
Status
Live at invest.forloopcowboy.com. – Experimental. Not financial advice.