Applied AI research & engineering
Truth Computing
We study how AI systems can reason, retrieve evidence, and act reliably in environments where mistakes are expensive.
We turn that research into working systems in law, healthcare, education, and other high-consequence domains.
A person stays in the loop on every decision the system makes.
By invitation. We take on a small number of engagements at a time.
Two sides of Truth Computing
Truth Computing builds systems, and explains the world they operate in.
Research & Engineering
We build and evaluate AI systems for domains where mistakes have real consequences.
Explore the labTruth Computing Media
Original reporting and field work on technology, cities, institutions, and the systems shaping everyday life.
Watch on YouTubeWhat we do
Research first, then a working system, then the hardening that lets you trust it.
We go inside a domain, find the question that is hardest to get right, and study it before we build. Different engagements move at different speeds, some are still in evaluation, others are already carrying real work.
Days
We sit with your team and find where the current system breaks.
Weeks to months
We build and evaluate a prototype against the failure cases we found.
Then
Where the evidence holds up, we harden it and put a human approval gate on every consequential action.
Live today
Two systems you can visit right now.
Clientlyy
Client communication infrastructure for personal injury law firms. Automates routine client updates while keeping attorneys in control of every message.
Visit clientlyy.com ›
Affordable Family Vision Center
Practice site and booking workflow built with our design partner in family optometry, Dr. Rosali Quintana. The broader clinical workflow behind it remains a prototype.
Visit affordablefamilyvision.com ›Technical capabilities
The categories of work behind the projects above, each linked to a real artifact.
01
AI Systems & Evaluation
Building and grading evaluation sets that test whether a retrieval system holds up against contradiction, poisoned context, and provenance traps.
02
Applied Machine Learning
Institutional research on evaluation and retrieval, alongside founder-led personal research on curriculum RL, model steering, and decision-making under uncertainty — kept clearly labeled by provenance.
03
Systems Engineering
Backend, workflow, and consent architecture for systems carrying real work — including the state/schema gate, send-path gate, and audit trail built for Truth Computing Health.
04
Reliability Engineering
Human approval gates, provenance evaluation, adversarial testing, tamper-evident audit trails, and explicit failure-state boundaries.
Truth Computing Media
Reporting and documentary work about technology, infrastructure, cities, institutions, and the people affected by them.
Independent field reporting from the same team building the systems.
Watch on YouTubeShipping log
Real, dated events. Nothing here is inferred.
-
Every figure on the By the Numbers page independently rechecked against its source; unsubstantiated figures removed
2026-06
This log only includes technical events with a real, sourced date from our own records; company formation and administrative/cosmetic updates are tracked elsewhere, not padded in here.
What we stand for
Human in the loop
You stay in control, always.
Our products keep a human in the loop by default. You decide what runs on its own and what waits for your approval. Full automation is available, and it stays off until you turn it on.
One concrete example, scoped to that system only: in Truth Computing Health's clinical workflow architecture, a message held in the CLINICAL_HOLD state can only be released by an optometrist (OD). This describes that system's architecture, not a claim generalized to every Truth Computing system — see the system record.
How we build
Built so you can check the work.
Trust
We show you the evidence behind what we tell you — see the research record and shipping log.
Reliability
Tested against contradiction, provenance traps, and poisoned context in the adversarial legal/crypto benchmark.
Privacy
Your data stays yours. Local-first design, where that is the design — see the privacy posture.
Security
We assume someone is trying to break it — the benchmark's poisoned documents are built to be mistaken for authority, and the system must refuse to cite them.
Auditability
Truth Computing Health's CLINICAL_HOLD transitions are logged in a tamper-evident, hash-chained audit trail — see the reliability architecture.
Matthew studied the systems, statistics, and theory this work rests on at Stanford.
Our team has worked on systems, product, and growth at Google, Microsoft, Stanford AI Lab, Stanford Medicine, and Synchrony.
The team
Our focus
Our attention belongs to the clients we serve.
We are currently working in education, healthcare, and law: three fields where a software mistake reaches a person directly.
How to read this site
Truth Computing publishes work at different stages, from early research to deployed products. Status labels reflect our current internal assessment, not a certification.
Some products and platforms described on this site are currently live or in production. Other descriptions may refer to planned features, capabilities, timelines, or future work. Unless specifically stated otherwise, forward-looking descriptions should not be understood as guarantees, commitments to deliver, certifications, or approvals. Availability and functionality may change as systems are tested and developed.
Industry statistics and third-party research are provided for context and do not represent Truth Computing performance or results.
Nothing on this site is intended as investment, financial, medical, legal, or other professional advice.
Specific products, systems, or research projects may include additional limitations or disclosures on their respective pages.
Spot an error? Email mtorre@truth-computing.com and we will correct or remove it.