Hackathon idea: Global AI- and FOSS-based ERP, tax and legal infrastructure
This is a project idea I wrote for the upcoming Cursor AI Hackathon, which will take place next weekend in Hamburg. See here for details. I had sent it into the hackathon’s Discord server. Let me know in case you’ll be there too :)
I want to co-found a new open source project: the global AI- and FOSS-based tax and legal infrastructure. It would not primarily be a tax or legal advisor, but the opposite: It would do everything for which no advice is needed, i.e. where the procedures are clearly defined by law and people working with it don’t actually need real intelligence, but only need to tediously and manually follow predefined procedures.
The vision is huge: creating one international project which supports all jurisdictions and all legal constructs, and allows interfacing with them not through tedious forms through a bunch of different websites or paper-based forms, but one globally harmonized interface based on human language input. The intention is to radically cut the costs of bureaucracy which are especially huge in European countries, by making the *business logic* of how bureaucracy works accessible to everyone, like the Linux kernel of business administration and paperwork, and to therefore minimize mental load experienced by many people when interacting with these formalistic systems. I believe that access to free high-quality bureaucracy is as important for a fair society as access to free high-quality health care. Bureaucracy is public infrastructure.
So of course, to first become able to handle tax-related data consistently, we first need to reflect the financial situation of an individual or a company accurately, i.e. we need to basically build our own ERP (Enterprise Resource Planning) from the ground up. The starting point will be nothing but self-hosted AI-based receipt and invoice parsing (this would be the MVP for the hackathon), which is something basically everyone with tax-deductible expenses needs, then writing invoices, handling purchase orders, revenue predictions etc. While there are already many commercial companies in this space, I think it’s time for us to build the “Linux kernel” of forms and institutional facts. If you’re philosophically inclined, also look up John Searle – who controls bureaucracy controls the people, so we need infrastructure so that everyone can control their own bureaucracy!
What I can contribute is being co-founder and CTO of German FinTech Divizend, where we’ve already been very successful since 2020; you might have heard of how we revolutionized withholding tax reclaims or of our mobile wealth app, the “Divizend Companion”, behind which I’m one of the main visionaries. We have legal entities in Germany, Luxembourg, Switzerland and Singapore.
Note that the intention of this is in no way a new commercial product of Divizend. Divizend is only my work background and where I got a lot of experience in what it needs to build enterprise-grade software, but my personal intention (as Julian) is to establish a new FOSS project to foster data autonomy in all financial, taxing and all other bureaucratic matters for everyone, to get away from personal data being stored in fragmented systems, towards always being “self-hosting first.”
Maybe you know DATEV or SAP, Lexoffice, sevDesk or TurboTax? They are the legacy corsets from which we want to liberate the people. No private person, company or any other legal entity, no matter where they are in the world, should need to pay for tax and accounting software ever again, of course except for when they want a managed service or actual advice. Economic accessibility is a global issue, not a national one.
Based on what I’ve researched so far, in terms of data formats we might want to use, there are XBRL Global Ledger, SAF-T (precise schema description here), UN/CEFACT and UN/EDIFACT, and ISO 9735. More on the legal side, there’s OpenFisca. In Germany, there’s DSFinV-K. Look them up if you’re interested to join the team, they are just my main research results so far; I’m personally quite new to ERP systems. For storing evidence documents (i.e. the foundation of bureaucracy, from which all other facts are derived), we might want to use MinIO or something equivalent.
As the architecture, we might consider an event-driven/Kappa architecture to enable optimal traceability and auditability, with a streamstore (e.g. Kafka, Redpanda, S2) as the single point of truth, plus arbitrary materialized views (e.g. using ksqlDB), to enable real-time by default. Or do you know about enterprise compliance, including ISO 27001 and SOC 2 Type 2? Apart from choosing a universal exchange format, we’ll later need convertability into local formats like German XRechnung. I had also discovered Nume, which calls itself the “AI CFO”, where we might get some UI inspiration from.
Happy to exchange ideas before the hackathon :) Please also not that I’m not participating in this hackathon to win. If there’s anything that disqualifies me for whatever reason, so be it. I don’t care about your rules. I also do not care about investors at all. I’m doing this because I think humanity needs it and because I want to find nice people to work with. I’m primarily process-oriented, not outcome-oriented, i.e., I care about us having fun while doing it.
Contact:
julian@hermesloom.org