Why NIH is a natural fit for our ML stack
Precision Federal is pursuing opportunities at the National Institutes of Health. NIH runs the largest non-defense SBIR program in the United States — roughly $1.3B across SBIR and STTR awards annually — and allocates that capital across 27 institutes and centers spanning every major disease area, biological system, and life stage. For an AI/ML small business with a production federal health ML system already shipped at SAMHSA, the NIH ecosystem is the deepest pool of AI-friendly solicitations on the civilian federal side.
Our founder is a Kaggle Top 200 data scientist whose federal past performance anchor is a production ML system at the Substance Abuse and Mental Health Services Administration (an HHS sister agency to NIH). That work proves three things NIH program officers and Scientific Review Officers routinely look for: we can handle sensitive federal health data, we can get a model through full security review and ATO, and we can operate a model in production beyond a proof-of-concept. See the documented SAMHSA case study and the broader past performance page.
All 27 institutes, plus CIT — where the AI/ML topics concentrate
NIH is not a monolith. The 27 institutes and centers each run their own SBIR solicitations, programmatic priorities, and scientific review groups. Our targeting map:
- NCI — National Cancer Institute. The single largest SBIR budget line at NIH. Cancer imaging AI (pathology WSI, radiology), genomics ML, digital twin oncology, clinical trials matching, SEER data analytics.
- NIAID — National Institute of Allergy and Infectious Diseases. Infectious disease modeling, HIV and pandemic preparedness ML, vaccine development informatics.
- NHLBI — National Heart, Lung, and Blood Institute. Cardiovascular imaging ML, cohort analytics (BioLINCC), wearables and continuous signal ML.
- NIMH — National Institute of Mental Health. Mental health ML, clinical risk stratification, digital phenotyping, SAMHSA-adjacent data work.
- NIDA — National Institute on Drug Abuse. Substance use modeling, opioid analytics, HEAL Initiative data. Directly adjacent to our SAMHSA past performance.
- NIA — National Institute on Aging. Alzheimer's and dementia ML, longitudinal cohort analytics, ADNI-adjacent imaging pipelines.
- NHGRI — National Human Genome Research Institute. Genomic ML, variant calling and interpretation, rare disease prioritization, dbGaP-scoped work.
- NINDS — National Institute of Neurological Disorders and Stroke. Neuroimaging AI, stroke decision support, EEG and iEEG ML.
- NIBIB — National Institute of Biomedical Imaging and Bioengineering. Medical imaging foundation models, device-embedded ML, federated learning.
- NLM — National Library of Medicine. Biomedical NLP, PubMed and MEDLINE AI, biomedical informatics, literature-scale LLMs and RAG.
- NIDDK — Diabetes, Digestive and Kidney Diseases. Continuous glucose monitor ML, endoscopy AI, kidney outcomes modeling.
- NICHD — Child Health and Human Development. Pediatric ML, rare disease, maternal health.
- NEI — National Eye Institute. Ophthalmic imaging AI (OCT, fundus).
- NIDCD — Deafness and Communication Disorders. Audio ML, speech and hearing analytics.
- NIDCR — Dental and Craniofacial Research. Dental imaging AI.
- NIEHS — Environmental Health Sciences. Exposomics, environmental ML.
- NIGMS — General Medical Sciences. Sepsis ML, pharmacology modeling.
- NIAAA — Alcohol Abuse and Alcoholism. Alcohol use modeling, behavioral health overlap with SAMHSA.
- NIAMS — Arthritis and Musculoskeletal. Imaging and biomarker ML.
- NIMHD — Minority Health and Health Disparities. Equity-centered ML, bias audits.
- NINR — Nursing Research. Clinical workflow and nursing informatics ML.
- NCCIH — Complementary and Integrative Health.
- NCATS — Advancing Translational Sciences. Drug repurposing ML, clinical trials innovation, rare disease informatics.
- FIC — Fogarty International Center. Global health data.
- CC — NIH Clinical Center. Intramural clinical ML partnerships.
- CSR — Center for Scientific Review. Scientific review operations (agentic workflow opportunity).
- CIT — Center for Information Technology. The NIH IT backbone. Network, HPC (Biowulf), identity, and shared services modernization. High-relevance for cloud and data engineering scope.
NIH SBIR and STTR — the mechanics that matter
NIH SBIR and STTR are the cleanest AI/ML pathway into the agency for a small business. Key mechanics for firms at our registration stage:
- Three standard receipt dates per year — typically early September, early January, and early April. RFAs often add one-off dates with narrower topic scope.
- Phase I — up to $306,872 standard cap, 6 to 12 months, feasibility scope. Topic-specific waivers can exceed the cap.
- Phase II — up to $2,045,816 standard cap, 24 months, prototype and demonstration. Waivers again available.
- Direct-to-Phase-II — available on many NIH SBIR announcements for firms with demonstrated Phase I equivalent work. Our production SAMHSA ML system is the kind of artifact that supports a D2P2 narrative.
- Phase IIB and commercialization supplements — NIH has the most developed post-Phase-II continuation ecosystem in federal SBIR.
- Fast-Track — combined Phase I and Phase II submission.
Submissions flow through Grants.gov but administrative tracking, reviews, just-in-time requests, and award management all happen in eRA Commons. Firm registration, Program Director / PI registration, and Signing Official setup in eRA Commons are prerequisites. We have the SAM.gov, UEI, and SBA small business foundation already in place, which is half the eRA Commons setup.
The foreign-ownership rule — and why it does not block us
NIH SBIR eligibility is governed by 13 CFR 121.702. The default rule: the small business must be more than 50% directly owned and controlled by one or more individuals who are U.S. citizens or permanent resident aliens, or by other small businesses that are themselves U.S.-owned. A narrow exception allows majority ownership by multiple venture capital operating companies, hedge funds, or private equity firms on specific NIH announcements, but not all.
Precision Delivery Federal LLC is U.S.-owned and controlled — the founder is a long-term U.S. permanent resident operating from Ames, Iowa, with the company wholly held. We meet the default ownership threshold. We do not rely on the VC-majority exception. This matters because some NIH SBIR topics explicitly exclude VC-majority firms, and our ownership structure is compatible with every NIH SBIR announcement category.
Biomedical AI scopes where our stack is strongest
Biomedical LLM agents and evidence synthesis
RAG-backed agentic systems over PubMed, ClinicalTrials.gov, and NIH RePORTER. Grant review triage, literature-at-scale synthesis, systematic review acceleration. See Agentic AI.
Genomics and variant ML
Variant calling, annotation, and interpretation pipelines. dbGaP-compliant controlled-access workflows. Rare disease gene prioritization. Integration with ClinVar and gnomAD.
Medical imaging AI
Pathology whole-slide, radiology, ophthalmic OCT. Foundation model fine-tuning, federated learning where data cannot centralize, model cards and bias audits.
Clinical trials matching and recruitment
EHR phenotyping, eligibility extraction, trial-patient matching ML. NCATS and NCI-relevant.
Pharmacovigilance and safety NLP
Adverse event signal detection over FAERS, clinical narrative NLP, structured extraction. Cross-relevant to FDA CDER.
Cohort and EHR phenotyping
All of Us and NIH-funded cohort data analytics. Phenotype algorithm development, temporal modeling.
CIT infrastructure modernization
Biowulf-adjacent HPC, STRIDES cloud migration, identity and data governance. See Cloud Architecture and Data Engineering.
Grants.gov, eRA Commons, and our registration stack
The administrative surface for NIH funding looks intimidating from the outside. In practice it is five systems that must be aligned: SAM.gov (UEI and CAGE), Grants.gov (application submission), eRA Commons (tracking and award management), the NIH Commons (PI profile), and the firm's own DUNS-era transition to UEI. Our registrations:
- SAM.gov — ACTIVE. UEI Y2JVCZXT9HP5, CAGE 1AYQ0, renewal March 29, 2027.
- SBA Company Registry — in progress for SBIR firm registration.
- eRA Commons — firm and PI registration on the planning board ahead of our first NIH submission window.
- Grants.gov — organization registration tied to SAM.gov UEI.
- Research.gov (NSF) — already complete in parallel for NSF SBIR.
Controlled-access data: dbGaP, GDC, All of Us
NIH AI work routinely touches controlled-access data. Each has a distinct governance layer we design around:
- dbGaP — Database of Genotypes and Phenotypes. Data Access Committee review, DUC agreements, encryption-in-transit and at-rest, approved environment attestation.
- GDC — Genomic Data Commons. Controlled-access tiers with token-based access and audit logs.
- All of Us Researcher Workbench — Terra-based workbench with enclave execution, no data exfiltration.
- SEER-Medicare linkage — data use agreements with NCI and CMS.
This is the same governance discipline we applied at SAMHSA under 42 CFR Part 2. We know how to build ML that lives inside a controlled-access environment.
Teaming: academic partners and NIH-experienced primes
STTR requires a formal university partnership (30% minimum to the small business, 30% minimum to the research institution). SBIR does not require a partner but often benefits from one on the science. We team with:
- University PIs with active NIH R01 or program-project grants in AI/ML-relevant domains.
- NIH-experienced primes for larger BAA, IDIQ, and task-order work at CIT and institute-specific CIOs.
- Clinical institutions willing to host validation studies on ML outputs.
How to engage on an NIH requirement
Three entry points work:
- SBIR or STTR teaming — you have a topic in mind (standard date or RFA) and need an AI/ML small business with ATO experience.
- CIT or institute IT subcontracting — you are a prime on a CIT or IC-specific vehicle and want an AI specialty subcontractor.
- Direct inquiry — a program office wants a small AI firm with federal health past performance.
Email [email protected]. We respond within 24 hours with a fit assessment, a rough level of effort, and a teaming construct.