Jordan Etzig

That's Collaborative Intelligence. That's the second thing RNDA does. And it might be bigger than the first. rnda.io

What if the data just didn't exist? Encode it. Delete the raw data permanently. Keep the signature. Now you can share the intelligence without sharing the data. No liability. No exposure. No walls needed.

The most valuable datasets in the world can't be shared. Not because of competition. Because of liability.

@andrewchen The GPU is the new "I'll just have one drink"

finding the main downside with experimenting with local AI models is that you end up buying one GPU, then another, then another, then another… But I’m running qwen3.6 27b dense at 100 tok/s now on a 5090 eGPU! It feels like sonnet 4.6? Fast and highly usable I figure the GPUs I have will now increase in value over the next few years so it’ll all be worth it

@eurie_kim That's exactly what we built. Encode it, delete the raw data permanently, keep the signature. The data is gone — the intelligence stays. Trust isn't a moat when you store less. It's a moat when you store nothing.

@jordan_etzig this is the part nobody talks about enough. the layer closest to the human generates the most sensitive data, and most companies haven't figured out how to learn from it without holding it permanently. that's actually a big part of why trust becomes a moat.

every acceleration moment looks the same in hindsight: 2007: mobile. everyone debated whether the iPhone would matter. then instagram, uber, doordash. 2012: social/mobile. "is this a platform?" then came a generation of consumer companies that redefined how people live. now: AI. and once again, the biggest outcomes won't only come from the best models. they'll come from who builds the product millions of people reach for without thinking. the pattern is always: infra settles → application layer explodes → the companies that win are the ones closest to the human.

That's the architecture we built. That's RNDA. New category. Multiple patents filed. 31 data types — from medical records to financial transactions to genomics. The assumption everyone made was wrong. We built the proof.

Building something with no comparable is a strange experience. No "it's like X but Y." No category to fit into. No benchmark to point to. Just a problem that's been unsolved for 50 years.

The database assumption: reliable records require retaining the raw values. What if the signature is more stable than the value? Encode at ingest, discard the original. The semantic record persists. The raw data doesn't. Databases endure. The question is what they actually need to store.

the whole point of databases is to keep reliable records of values that change over time. in general, they can be trusted to be stable over time. the whole point of LLMs is to raise money. in general they cannot be trusted. the former (databases) will endure forever; the latter (LLMs) will eventually be displaced by something more stable and efficient.

We've built RNDA which could be useful depending on what stage and data volumes you're working with — encoding protein sequence and binding data to compressed signatures could help manage the scale of variant screening while keeping proprietary research protected. For biological sequence data — protein sequences, binding results, variant libraries — RNDA encodes each record to a fixed 256-byte binary signature and permanently discards the original. Raw files that are megabytes become 256 bytes. On genomic sequence data we've measured compression ratios of 140,835x. The signatures support semantic similarity search via a discrimination gap metric. A gap near 1.0 means the system reliably distinguishes similar sequences from dissimilar ones — so you can query "find candidates with similar binding profiles to this sequence" across millions of encoded variants without ever reconstructing the raw data. For multi-lab collaboration: each lab encodes their proprietary data locally. The signatures can be shared and queried across institutions. The raw research never leaves the lab. Happy to show you what this looks like on real biological data.

Planning to make antibodies for Ebola. Anyone interested in collaborating.

Applied to the speed run. I'm a solo founder from Las Vegas — built RNDA, the first data architecture where raw data is permanently discarded at ingest (not encrypted, gone), while preserving full semantic queryability. Makes data breaches of sensitive data structurally impossible. 31+ domains validated, multiple patents filed.

@Trace_Cohen Token minimaxing at the model layer. We built RNDA of data minimaxing at the infrastructure layer. 256-byte signatures instead of raw files. Full semantic queryability. The original discarded permanently. Deceptively lightweight is the right framing — for data too.

“Token maxxing” is mostly performative. It is not tied to real business metrics, real productivity metrics, or even necessarily better outcomes. Right now parts of the AI ecosystem are treating token consumption like horsepower numbers on a car: bigger context windows, longer chains, more agents talking to more agents, infinite loops generating infinite text. But consuming more tokens is not inherently intelligence. It is not inherently efficiency. And it is definitely not inherently value creation. If anything, the long-term direction probably looks more like token minimaxing: getting the maximum amount of useful work, reasoning, coordination, and output from the minimum necessary compute and token usage. The best systems historically do not win because they consume the most resources. They win because they achieve the best outcome per unit of resource. The future enterprise KPI will probably not be: “How many tokens did your system use?” It will be: “How much operational leverage did this create relative to cost, latency, reliability, and human oversight?” The truly impressive AI systems will likely feel deceptively lightweight. Small prompts. Minimal orchestration. Focused context. Efficient reasoning. High signal density. Strong outcomes. Not infinite chains of AI agents generating 400 pages of recursive slop to complete a task a great operator could have solved in three decisions.

I'm a founder from Las Vegas — built RNDA, the first data architecture where raw data is permanently discarded at ingest (not encrypted, gone), while preserving full semantic queryability. Makes data breaches of sensitive data structurally impossible. 31+ domains validated, multiple patents filed.

Full article: reviewjournal.com/business/las-v…

Las Vegas is projected to lead the nation in tech employment growth in 2026. Not Silicon Valley. Not Austin. Las Vegas. Building RNDA here. The infrastructure that makes data breaches structurally impossible — filed from Las Vegas, validated across 31+ data types. The future of tech doesn't have one zip code.

Part of the GPU crunch is wasted compute on bloated data pipelines — querying raw files that are orders of magnitude larger than they need to be. We developed RNDA exactly for this reason. Encode at ingest, discard the raw data. 256-byte signatures query in milliseconds. The H100s you do have go a lot further.

GPU shortage is worse than ever. H100s cost more today than they did 3 years ago, and you cannot get them on-demand. The big AI labs have locked up most of the supply for years. I’m worried university researchers and individual developers simply won’t be able to get GPUs.