Complexity Labs is a research-led engineering company. We build the data platforms and AI systems that hard problems need, and we run the research those systems turn out to require. Our partners are in academia, industry and government.
What we doData platforms, document automation and machine learning, running inside your own cloud tenant and extending the systems you already have.
Learning theory, post-quantum cryptography and sublinear algorithms. We pick the questions whose answers change what we can ship.
Lineage, audit logs, and a person at every consequential step. Oversight and records requests are a design input, not an afterthought.
Your data sits in systems that were never meant to talk to each other.
One queryable, AI-ready surface over the whole estate: files, object stores, databases, warehouses, streams and APIs. Nothing is replaced. You declare where data lives, how it transforms and where results land, and every change propagates on its own.
How NRV worksThe document backlog grows faster than anyone can read it.
Agents extract what matters from filings, forms and archives, check it against your rulebook, and draft the next action with citations to both the document and the rule. A person approves every consequential step, and every step is logged.
How Machina worksPublic work has to be explainable to people who were not in the room.
Data platforms, document automation and AI for agencies, built to be examined. We deploy inside your tenant, integrate through APIs, and leave your source systems unmodified. Registered, SAM.gov active, with a capability statement you can print for your files.
Government servicesProduction systems cannot stop while you modernize them.
We unify estates scattered across decades of systems and rebuild aging pipelines while they keep serving traffic. Lineage, quality scores and access controls come from day one, because retrofitting them onto a live platform is the expensive way.
Industry servicesBuying a model is easy. Checking that it learned what it claims is not.
The theory side works on PAC learning and, in particular, verification: what it takes to check a model with less work than training it yourself. The applied side ships bias testing under the NIST AI Risk Management Framework, drift detection in production, and retrieval that cites its sources.
Machine learning researchEverything encrypted today can be recorded today and read the day a large quantum computer exists.
We audit where you are exposed to harvest-now-decrypt-later collection, then migrate: key exchange to ML-KEM (FIPS 203) and signatures to ML-DSA (FIPS 204). We build on algorithm-agnostic libraries, so your next migration is a configuration change. The research arm works on lattice problems and verifiable audit trails.
Cryptography researchSome datasets are too large to read end to end.
Sketches and streaming estimators answer anyway, and state how much to trust the answer: Morris counters, count-min sketches, locality-sensitive hashing. You choose the accuracy and the confidence, and the algorithm chooses how little it can get away with reading. These primitives end up inside NRV.
Sublinear algorithms researchThe work looks different depending on which constraints are the real ones. Pick yours, and the rest of the site follows you there.
Shared problems and joint work in learning theory, complexity and cryptography, plus data infrastructure that keeps results reproducible.
Regulated operations that cannot stop: estates unified, pipelines rebuilt in place, and audits that are a query rather than a quarter.
Agencies and their primes: procurement-ready, audit-ready, and deployed inside your own tenant.
One address, read by the founder. If it is intractable, so much the better.
Write to us