Hi HN, I made a simple single-page tool that computes portfolios return / performance (TWR and XIRR), gross and net of fees and tax. It requires logging all transactions. It's client-side, with no accounts, and data lives in the browser localStorage.

It started as a personal tool . I wanted a way to understand what was the return of my portfolio and of a tenanted rental property (XIRR).

The experimental part is an import: you upload a broker's trade notice PDF and an LLM proposes one ledger row, which you confirm. It solves a problem I had: understanding my bank's documents, especially for bonds, at least here in Italy, where there are multiple numbers to add or subtract to get the real return.

It's an alpha. I've only tested it on real documents from one Italian bank (about 40 notices, all correct on the first attempt) and on a few synthetic ones from other countries, so I'd really like to hear how it does on yours.

Caveats: the PDF goes to the model provider and is neither stored nor logged by me. The AI import needs an access code, because I'm paying for the API calls out of pocket, so ask me by writing to hello@returnledger.site and I'll send one. The tracker itself works without it.

I'm looking for some feedback regarding AI interpreting other brokers' documents (no need look at sensitive information, though).

I'm happy to answer questions about the validation approach or the architecture (Azure Static Web Apps with a small Python function behind it).