tav1

I built a bank statement converter that uses Intel TDX enclaves to process PDFs in hardware-encrypted memory. The key difference: neither I (the service provider) nor the cloud host can access the data being processed. The problem: Every bank statement converter requires you to trust that they won't look at your financial data. Even "we don't store it" is just a policy promise. The approach: Intel TDX creates isolated execution environments inside the CPU where:

Data is processed in encrypted memory The service provider literally cannot access the memory contents Remote attestation provides cryptographic proof the code is unmodified Data exists only during processing (seconds), then RAM is wiped

Why it matters:

Forensic accountants processing divorce/litigation cases Mortgage lenders doing income verification Anyone who doesn't want their net worth visible to a SaaS provider

Honest question for HN: Is this overkill, or is "trust us" fundamentally broken for financial document processing?

When you say "we process your data, we don't store it" does the PDF get sent to a third-party OCR service that might retain it temporarily or use it for training? I'm building something similar using Intel TDX enclaves where processing happens in hardware encrypted memory that literally can't be accessed. Curious if users actually care about that level of privacy or if "deleted in 24hrs" is enough.

Live Demo: joinmesh.xyz

Hi HN,

I built Mesh because I wanted a self-hosted alternative to Slack that didn't require a complex backend or central database. It runs entirely on the Nostr protocol using the NIP-29 specification (relay-based groups).

The architecture is:

Backend: A custom NIP-29 compliant relay written in Rust.

Frontend: Next.js client that connects via WebSockets.

Auth: No email/password; uses end-to-end cryptographic keys (Nostr identity).

The goal was to prove that you can have "Slack-like" real-time features (channels, admin roles, kick/ban) without giving up data sovereignty. Since it's Nostr, you can swap the relay out anytime and migrate your data easily.

It’s still early days (v1), so I’d love feedback on the Rust relay implementation or the group handling logic.

Happy to answer any questions!