Over the last few months I've been building a prototype around a question I can't stop thinking about:
How do you know when an AI-generated financial claim is actually trustworthy?
The obvious answer is "use a better model."
The more I've built, the less I believe that's the real solution.
The harder problems have turned out to be things like:
- representing evidence across multiple documents
- reconciling conflicting financial values
- deterministic rule evaluation
- calculation traceability
- versioned verification logic
- deciding what can actually be verified versus what should remain outside scope
It's less of a chatbot problem and more of a systems, data, and engineering problem.
That's exactly why I enjoy working on it.
I'm still at the prototype stage, but every week the project becomes less about prompting LLMs and more about building infrastructure for trustworthy AI.
If you're the kind of engineer who gets excited by:
- C++
- distributed systems
- compilers
- formal methods
- financial systems
- document intelligence
- verification
- evaluation
- deterministic software
I'd genuinely like to hear what kinds of problems you're working on.
Not recruiting today.
Mostly looking to meet people who enjoy building difficult systems.
Some of the best opportunities I've had started as technical conversations rather than interviews.
[link] [comments]