AI-driven suicide risk prevention that identifies the root causes of veteran depression and anxiety — and connects them to local resources that match their individual profile. 100% anonymous. Zero barriers.
The Department of Veterans Affairs purports universal screening for veteran suicide risk. But a finer examination reveals a glaring caveat: only veterans who engage with in-person VA care are actually screened.
For many reasons — including the VA being embroiled in numerous public scandals since at least 2014 over neglect, long wait times, and poor services — roughly 6.84 million veterans have never engaged with the VA at all and, by extension, were never screened.
Worse still, Department of Defense data indicates that suicide prevalence is highest within this unscreened population — the very group the VA has deemed "forgotten."
Current approaches are passive: they wait for veterans to come to them. Veterans face additional barriers including warrior culture that suppresses help-seeking, distrust of the VA system, fear of career repercussions, and the stigma of admitting vulnerability.
of all US veterans are never screened by VA's Risk ID system. Out of roughly 18 million veterans (2020), that means over 6.84 million fall outside the safety net entirely.
U.S. Census Bureau & Vespa, 2020; Wang et al., 2021The rate of veteran suicide is more than double that of civilian counterparts, according to the VA's 2023 National Veteran Suicide Prevention Report.
VA Office of Mental Health and Suicide Prevention, 2023of individuals experiencing depression or suicidal ideation had not received treatment — driven by stigma, lack of confidentiality, and fear of repercussions.
Garlow et al., 2008increase in the US suicide rate between 1999 and 2019. In 2020, suicide was the 12th leading cause of death overall — and as high as the 2nd among those under 44.
Hedegaard et al., 2020; NIMH, 2022A veteran walks into a county office, a food bank, a shelter, a 988 call center, or a DMVA desk. Instead of a generic pamphlet, they receive a tailored set of local recommendations — built from their interests, military service, and (if they choose) a full depression and anxiety screen — all in under two minutes.
A credentialed staff member — counselor, social worker, volunteer, 988 operator — opens ROOTS on any tablet or phone. The veteran is asked up to three optional questions.
No name. No SSN. No date of birth. No personally identifiable information. Just a zip code to find local help. The system produces a synthesized report of nearby nonprofits matched to the individual's needs and interests — with plain-language explanations for every recommendation, so counselors can verify the logic on the spot.
This isn't another pamphlet. It's a custom-built referral package, validated against IRS Exempt Organizations data, filtered by NTEE category and organizational capacity, and tailored to who this specific veteran is — not who veterans are on average.
Every output is scored by a second AI judge against standards co-developed with mental health clinicians. The system grades itself, surfaces low-quality runs for review, and improves continuously. Counselors stay in the loop. Veterans stay anonymous. The right help finds the right person.
ROOTS uses AI to identify the root causes of each veteran's depression and anxiety — individually, not generically — and connects them to validated local nonprofits. 100% anonymous. No PII required. All questions optional.
Deployable at DOD discharge, VA, NGO, shelter, or 988 line — closing the gap on the ~38% of veterans never reached by VA Risk ID. This single intervention closes the gap on roughly 6.84 million unscreened veterans.
A mobile-ready screener (iPhone or tablet) uses machine learning models with AUC-ROC scores of .942 (depression, XGBoost) and .948 (anxiety, LightGBM) to detect conditions — then identifies the specific drivers behind each veteran's distress using explainable AI (SHAP values).
Moves beyond generic population insights to individualized ones. Not "many veterans struggle with sleep disturbances" but rather "this veteran is anxious because they have been struggling to meet their family's financial obligations." Individual roots, not population averages.
Considers pre-service hobbies, military branch and experience, and personal interests. Synthesizes all details together to build a complete individual profile that informs which resources will actually resonate with this specific veteran.
Retrieval Augmented Generation searches any US zip code, filters IRS-validated nonprofits by NTEE code and size, and matches them to the individual's profile — either as volunteer opportunities or as recipient services. Reduces strain on overburdened governmental systems while connecting veterans to community organizations that align with who they are.
Every output is scored by a second AI against standards co-developed with clinicians. The system judges itself and surfaces low-scoring runs for continuous improvement. A 988 operator in Shreveport can instantly find validated NGOs in Eugene, Oregon — 2,200 miles away — tailored to that caller's specific needs, with quality assured at every step.
A first-of-its-kind whole-of-community solution poised to drive down instances of veteran suicide in the United States.
Anonymous, no PII. All questions are optional. Removes the stigma, security-clearance fear, and institutional distrust that block veterans from seeking help in the first place.
Every AI step shows its reasoning. Counselors and clinicians verify each recommendation before handing it to the veteran. Local SHAP values explain precisely why the system reached its conclusions.
Uses SHapley Additive exPlanations to identify why this specific veteran is struggling — financial stress, housing instability, family obligations — not generic population-level trends.
Brings help to where they are — food banks, shelters, county offices, 988 lines, religious organizations — not where we assume they'll go. A whole-of-community safety net.
Cross-referenced, high-quality referrals in seconds. Lets mental health professionals focus on care, not on hunting for resources. Consistent quality regardless of geographic distance.
We embed with each deployment and build the interface around their operational reality — counselors, intake staff, volunteers — not the other way around.
We are not chasing a grant. We are building something that meets veterans where they are, rather than where they are expected to be — a proven, privacy-first, AI-driven screener that closes the 6.84 million-veteran gap and becomes a recurring budget line in the DMVA and beyond.
Put ROOTS in front of senior clinical and research leadership — PhDs, PsyDs, and MDs — and invite them to poke holes, stress-test assumptions, and rally around the methodology as co-architects.
Our team embeds with customers. We build the UI, workflow, and outputs to whatever shape the operator needs for operational success — sitting side-by-side with frontline intake staff.
Move from pilot to recurring state budget line. ROOTS becomes a funded item within the DMVA or an adjacent bureau — ratified by the Pennsylvania legislature.
Pennsylvania becomes the proof case. The architecture scales to every state, federal agencies, schools, and the general public — ratified at the Congressional level.
ROOTS is grounded in doctoral research successfully defended at the George Washington University School of Engineering and Applied Science, and reviewed by members of industry. The patent and intellectual property are held by Nicholas C. Birosik.
14 candidate models were evaluated using PyCaret's soft Auto-ML framework. Superior-than-baseline XGBoost (depression, AUC-ROC .942) and LightGBM (anxiety, AUC-ROC .948) models were identified, tuned, and implemented. Statistical significance was confirmed via McNemar's contingency test against baseline logistic regression.
A novel method called Dispersion-Penalized Convex Mesh (DPCM) was developed for Model Class Evaluation — a new contribution to the machine learning literature for ranking and visualizing the predictive power of grouped models.
Local SHapley Additive exPlanations (SHAP) perform root-cause analysis at the individual level, shifting focus from global patterns to personalized insights. Global SHAP bee-swarm diagrams revealed previously undocumented population-level patterns including compelling evidence for data seasonality and distribution shifts in younger veteran demographics.
The system leverages a RAG corpus of IRS Exempt Organizations data covering 50+ NTEE categories, paired with a haversine distance matrix between all US zip codes for rapid geographic subsetting. Regular expressions match veteran interests, military context, and identified needs against NGO profiles to deliver targeted, actionable referrals.
A distributed, GPU-enabled Python backend processes requests through a chained series of Large Language Models. Web-based frontend accessible on any device. Designed for horizontal scaling via containerization to decrease latency and increase throughput for nationwide deployment.
ROOTS can be deployed by any organization with incidental veteran contact — extending the safety net far beyond VA walls.
Implement at military discharge to identify high-risk veterans before they leave service — achieving universal screening and providing proactive referrals for healthy reentry into civilian life.
A call-line operator in any city can instantly find validated, local NGOs near the caller — regardless of distance — tailored to their specific root causes, interests, and military background.
Any frontline worker who encounters a veteran — at a food bank, shelter, religious organization, or outreach program — can administer the free, low-barrier screener on a tablet and provide life-saving referrals on the spot.
Introduced by Congressman Ryan Mackenzie, H.R. 8486 directs the VA to fund AI-driven tools that predict which veterans are most at risk of suicide and connect them with help. The National Mental Health Network is an active proponent of this legislation and has submitted technical recommendations to strengthen its safeguards — because building an AI model is the beginning, not the end.
Risk factors shift with the time of year — holiday loneliness, post-discharge vulnerability windows, economic cycles. A model trained on one snapshot will miss these patterns. We recommended time-aware feature engineering and rolling retraining windows be written into the bill.
Over the bill's three-year funding window, the world changes — VA policies, economic conditions, new veteran cohorts. Without mandated monitoring, a model degrades silently. No error message, no red light — just quietly wrong predictions. We proposed quarterly drift-detection pipelines and minimum retraining thresholds.
We recommended four safeguards: mandatory model performance check-ups, individual-level explainability (the AI shows its work), demographic fairness auditing across all veteran populations, and a human-in-the-loop requirement — the AI supports the clinician, never replaces them.
Full technical analysis of prediction seasonality, model drift, and proposed legislative guardrails for AI-driven veteran suicide risk prevention systems.
A non-technical walkthrough of the same recommendations — what seasonality and model drift mean in plain English, and why these safeguards matter for veterans.
Both documents were prepared by Nicholas C. Birosik, D.Eng. and submitted on behalf of The National Mental Health Network in support of H.R. 8486. The recommendations are offered in the spirit of strengthening an already commendable piece of legislation.
The fight to close the 6.84 million-veteran gap crosses party lines, levels of government, and sectors. A Democratic state senator and a Republican U.S. congressman have both formally endorsed The National Mental Health Network and ROOTS — proof that veteran suicide prevention is not a partisan issue. Below are the leaders who have signed letters of support and the stakeholders we are honored to be in active conversation with.
The National Mental Health Network meets the critical infrastructure role as it provides up to 6.2 million veterans with timely access to mental health support. I fully recommend approval of The National Mental Health Network as a Qualified Critical Infrastructure NGO under 15 U.S. Code § 7442.
The National Mental Health Network's ROOTS program serves as critical infrastructure supporting our community's veteran population by providing accessible mental health screening and connection to local resources. This platform addresses a documented gap affecting 6.2 million veterans not currently served by Department of Veterans Affairs screening capacity.
Stakeholders engaged, scheduled, or actively being briefed as part of the PA legislative pathway.
Draft letter language is ready for your review and signature — adopt, edit, or use as a starting point. Letters of support accompany our applications for federal grant funding and signal alignment with Commonwealth and federal priorities in veteran care.
Documentation supporting the ROOTS rollout, letters of support, and our legislative briefing materials for H.R. 8486 are available below. For additional materials — including the full doctoral praxis, clinical methodology brief, or a custom presentation for your office — please reach out directly.
The full leadership-review playbook: four-phase strategy, current engagement status, and a plain-terms walkthrough of the program from the veteran's perspective.
The first signed letter of support, recommending NMHN as a Qualified Critical Infrastructure NGO under 15 U.S. Code § 7442. Signed April 30, 2026.
Congressional letter supporting NMHN's designation as a Qualified Critical Infrastructure NGO. Cites the ROOTS platform's role in closing the veteran screening gap. Signed May 27, 2026.
Technical analysis of prediction seasonality, model drift, and proposed legislative guardrails for AI-driven veteran suicide risk prevention under H.R. 8486.
Non-technical walkthrough of the safeguard recommendations — what seasonality and model drift mean in plain English and why they matter for veterans.
Full Finding the Forgotten doctoral praxis, defended at GWU. Includes model evaluation, SHAP analysis, RAG architecture, and statistical validation.
30-minute working session for clinicians, legislators, agency leadership, or partner organizations. Walk-through of methodology, demo, and Q&A.
"Veterans are found. Veterans are seen.— Finding the Forgotten, Birosik (2025)
Veterans are understood. Veterans are healed."
The National Mental Health Network is a 501(c)(3) nonprofit organization, faithfully serving veterans under NicNac Charities Inc. Your contribution directly supports the deployment of ROOTS to reach the 6.84 million veterans who have never been screened.
If You or Someone You Know Is in Crisis
Call or Text 988
The 988 Suicide & Crisis Lifeline provides free, confidential, 24/7 support. You can also chat at 988lifeline.org.
For the Veterans Crisis Line, press 1 after dialing 988.
Disclaimer: ROOTS is a screening and resource-referral tool designed to support — not replace — the professional judgment of licensed mental health professionals, counselors, and medical providers. ROOTS does not diagnose, treat, or provide clinical care for any mental health condition. The information, referrals, and outputs generated by the system are intended to assist qualified professionals and should not be construed as medical advice, psychiatric evaluation, or crisis intervention. In an emergency, dial 911 or contact the 988 Suicide & Crisis Lifeline immediately. The National Mental Health Network, NicNac Charities Inc., Dartboard Systems, and Nicholas C. Birosik assume no liability for actions taken or not taken based on ROOTS outputs. Use of this system constitutes agreement to these terms.