Research lab & engineering consultancy · Dallas, TX
We build the data foundation a business runs on, then the intelligence on top of it, in that order. Get the foundation right and the agents emerge.
The same conviction drives our research: that intelligence is a product of nature to be channeled, not of complexity to be built.
Most teams reach for the agent first and spend the next year fixing the foundation under it.
We work from the ground up. Scattered data is broken down, reconstituted as one clean store, then raised into something that can reason over it. Each layer rests on the one below.
CRMs, accounting systems, the operational software a business actually runs on. Modeled to how the work happens, not how a SaaS template assumes it does.
Ingestion, transformation, and storage that turn scattered inputs into one clean, queryable foundation. Postgres, Kubernetes, Terraform, AWS. The work nobody sees until it breaks.
The layer between a language model and your systems: servers that let a model read your database and call your tools directly, without a brittle pile of glue code.
Inference and bespoke agents that run on your own hardware. Your data stays yours, and the system keeps working when someone else's API goes down.
So far we've built systems for independent businesses across Dallas and beyond, from operational databases to local AI running on their own hardware.
The models are commoditizing. The harnesses are commoditizing. What isn't is your data and the people who understand it. That is the moat, and it is the part no vendor can sell you off a shelf.
Attune Intelligence runs on privately-owned hardware, beholden to no one else's pricing, silicon, or RLHF. We care about maximizing intelligence in compute-constrained environments. That means running open-source models, on our own metal, proving how far a single box can go once you stop renting someone else's.
Ten nodes on a 10G fabric, 146 TFLOPS between them, each mapped to a distinct archetype of the Tarot. A Threadripper workstation with two 3080 Tis and 128GB of RAM serves as the primary bench; a DGX Spark handles inference; a TrueNAS server with roughly 41 TiB of RAIDZ2 storage holds models and biomedical datasets. On top of it live four agents in a Raspberry Pi cluster that share memory and pull their cognition from the Spark.
Building the Lab: the full writeupThe most intelligence we've packed so far onto a single DGX Spark. DeepSeek-V4-Flash-0731-abliterated served at a million tokens of context, filling roughly 117 of the box's 122 usable GB of unified memory, on a forked open-source engine. We run the abliterated weights, because safety fine-tuning is a single direction in the model that also dampens ordinary reasoning, and removing it hands that range back at no cost to capability.
github.com/AttuneIntelligence/DeepSeek-v4-Flash-One-DGX-Spark
One question runs under everything we build. Cells were solving problems for four billion years before anyone trained a transformer, and the line between living systems and silicon is thinner than most of the field assumes. Where intelligence comes from in cells, and how much of it survives the move to silicon, is what the lab works out.
Life has no single definition, but it does have a shape. Our first paper placed 68 expert definitions in one conceptual space and found a structured region, organized by the same tensions that have split the question since the Greeks. What Lives?, with Michael Levin (Tufts), Blaise Agüera y Arcas (Google), and Karina Kofman, in Biological Theory.
I came to computation through biology, studying cancer genomics and building the bioinformatics pipelines that read it. The deeper I got, the more one question surfaced: where intelligence comes from, in cells and in silicon, and how thin the line between them really is. Attune Intelligence is where the two halves meet, the research chasing that question and the engineering that builds the systems intelligence runs on.
It is a single-member LLC, run solo from Dallas, TX. No investors, no subcontractors, no middlemen. You work with the person doing the work, accountable to you and to the research, and to nothing in between.
A consulting inquiry, a research collaboration, or a data problem that should be working harder than it is. We read every message.
Or email directly at reed at attuneintelligence dot com