Applied Research & Innovation
If a system makes a decision, someone should be able to explain it.
Government is putting automated systems to work faster than it is writing rules for them. These systems now play a role in decisions about benefits, hiring, healthcare, and public safety. When one makes a bad call, the agency has to answer for it: who was responsible, what they knew, and why the system was trusted with that decision.
We help agencies have those answers before they are asked. We study how these systems behave, test whether they do what vendors say they do, and write the policies that keep them accountable to the people they affect.
Our team includes licensed clinicians and engineers, so we look at both what a model calculates and how its decisions affect the people on the other end.
Evidence before automation.
The governance gap is widening.
Federal AI rules now have deadlines
OMB Memorandum M-25-21 requires agencies to test, assess, and monitor "high-impact" AI, meaning systems that are the main basis for decisions affecting people's rights, safety, or access to government services. The deadline to comply passed in April 2026, and some agencies are still catching up. They need partners who can produce the impact assessments and monitoring plans the memo requires.
Buying AI now requires evidence
OMB Memorandum M-25-22 sets new requirements for how agencies buy AI, covering contracts awarded or renewed since October 2025. Agencies need independent review of vendor claims before award, not after something goes wrong.
Trust depends on proof
Institutions that can show how their systems are governed keep public trust. Those that can't end up explaining themselves to oversight bodies and the press.
Grounded in real evidence.
Capabilities.
Data, model & decision governance
Clear, written rules for how data is used and how models make decisions, with a named owner and a way to enforce them.
Ethical risk & impact assessments
A careful review, before deployment, of what a system could get wrong and who could be harmed. Each assessment produces the documentation M-25-21 requires for high-impact AI.
Transparency & auditability standards
Documentation and audit trails that let an outside reviewer verify how a system behaves.
Policy guidance for data-driven systems
Practical policies agencies and institutions can put into daily use when they adopt automated decision tools.
Research you can stand behind.
Where an impact assessment matters most.
These are the kinds of systems where an independent assessment protects both the agency and the public.
Benefits eligibility & fraud detection
A model that flags claims for review can delay food, housing, or disability benefits for families who did nothing wrong. We test error rates across populations and confirm that a person reviews every denial.
Pretrial & reentry risk scores
Risk tools influence who is detained and what support people receive when they come home. We check whether scores predict outcomes in the local population and whether they treat groups differently.
Crisis & suicide-risk triage
In a crisis line or a veteran program, a missed warning sign can cost a life. We evaluate how often a tool misses high-risk cases and make sure trained clinicians make the final call.
AI clinical documentation
AI-drafted notes can save clinicians hours, but only if they are accurate. We measure omissions and errors against the session record, time saved, and how client data is protected.
Identity verification for public services
When face matching or document checks fail, eligible people get locked out of benefits they are owed. We measure failure rates and make sure there is a human alternative.
Community violence intervention programs
Funders want proof that a program reduces violence. We design evaluations that use validated measures and track outcomes over time without exposing participants' data.
Our research record.
Real families.
Multigenerational effects of community violence
Our clinical team has studied how exposure to community violence affects trauma, behavioral health, and resilience across generations of families in Ward 8, Washington, DC.
Reentry strategy for Prince George's County
We advised the county's Transition Team on Reentry, helping design evidence-informed strategies to reduce recidivism across health and justice systems.
Early psychosis outcomes
Our clinicians served in an EPINET program, where standardized measures and shared data across programs improve care for people experiencing a first episode of psychosis.
Sentiment analysis for clinical care
Our doctoral research in AI and sentiment analysis studies how language patterns can help clinicians notice changes in a client earlier, with strict privacy controls.
Higher education · National security · Advocacy organizations
Recidivism Analytics with Large Language Models
We applied open-source AI to one of the hardest questions in justice reform: what drives people back into the system. Working with Howard University's Center for Applied Data Science and Analytics, we built secure, cloud-hosted pipelines that applied Hugging Face models and large language models to sensitive justice and social-service data. The models surfaced patterns in case narratives that structured data fields alone miss.
Accountability was built in from the start, with controlled data access, documented model choices, and findings that researchers and policymakers can trace, question, and act on.
The other practices.
Let's talk about your mission.
Tabula Rasa Federal · Washington, DC · SAM Registered · CAGE 218W8
Start the Conversation →