VA AI Tech Sprint Follow-Ons: How an SDVOSB Turns a Demo Into a VA Contract

What is the VA AI Tech Sprint?

It’s the VA’s open-challenge on-ramp for outside AI builders. The Tech Sprint is run by the VA National Artificial Intelligence Institute (NAII) in partnership with the Office of the Chief Technology Officer (OCTO) and the VHA Innovation Ecosystem, under the prize-competition authority at 15 U.S.C. § 3719 (the America COMPETES Act). Each round runs roughly 90 to 120 days, organized into tracks tied to specific VA problem statements, with a cohort of finalists (typically 20–25 from 100+ applicants), access to synthetic or de-identified VA data, VA subject-matter-expert mentorship, and a demo-day finale with cash prizes.

The detail that matters for procurement: under the COMPETES Act, the government may negotiate a follow-on contract with a challenge winner without re-competing, and entrants keep their intellectual property — the government can’t take an interest in it without written consent. So the sprint isn’t just marketing exposure; it’s a structured competition that can stand in for a FAR Part 15 source selection. The most consequential recent round was the 2023–24 “AI Tech Sprint for Documenting VA Clinical Encounters,” a $1 million prize competition that drew roughly 152 teams in its main track and produced the Abridge and Knowtex pilots.

How does a Tech Sprint demo actually become a contract?

Here’s the worked example, because it’s the whole argument of this page.

The VA launched the clinical-documentation sprint in late 2023, named 25 finalists in March 2024, and announced winners that May. In July 2024 the VA’s Strategic Acquisition Center published a notice of intent to sole-source pilot contracts to Tech Sprint winners. By mid-2025 the pilots were on contract: Abridge AI at roughly $5.37 million (a 12-month base plus four six-month options), awarded June 2025, and Knowtex, awarded in July 2025 — both firm-fixed-price, both sole-source, both citing the COMPETES Act challenge as the justifying competition. The sole-source justification said it plainly: Abridge was selected because it was a winner of “a rigorous and competitive evaluation process authorized under the America COMPETES Act.” The pilot went live at 10 VA medical centers in October 2025.

Then came the part that matters most for an SDVOSB. In March 2026, the VA awarded the production rollout — scaling ambient scribe from 10 sites to more than 130 medical centers — to Rise8, a service-disabled veteran-owned small business, partnered with Thoughtworks Federal. The AI vendors (Abridge, Knowtex) supply the commercial scribe software; the SDVOSB does the production engineering: deeper workflow integration, observability, vendor onboarding, ATO sustainment, and national scale. Rise8 also holds a seat on the SPRUCE IDIQ, one plausible vehicle for the task order. One Tech Sprint with a $1M purse spun up a program now worth tens of millions across the pilot and production phases — and the integrator’s seat went to an SDVOSB.

What other VA AI pilots have crossed the “valley of death”?

The scribe story is the cleanest, but it isn’t the only one. The VA runs a large and varied AI portfolio at different stages of maturity.

The AI-Driven eFax Fix (AIEFF) parses the more than 13 million community-care faxes the VHA receives a year; first deployed at the C.W. Bill Young VAMC in Bay Pines, Florida, it cut average per-document handling time by about 31%. REACH VET, the VA’s suicide-risk prediction model, has run in production since 2017 and has been used to identify more than 130,000 veterans at elevated risk, always with a human in the loop. On the productivity side, the VA’s internal generative-AI tool reports more than 95,000 users onboarded, over 70% reporting improved job satisfaction, and an average of two to three hours saved per week, and the VA’s GitHub Copilot deployment serves more than 2,000 developers reporting roughly eight hours saved weekly. On the benefits side, machine-learning decision support helped bring the disability claims backlog under 100,000 in early 2026 for the first time in years.

The scale behind all of this: the VA’s 2025 AI use-case inventory lists 367 use cases, a large share of them classified as high-impact under OMB M-25-21 — the largest AI footprint in the federal government. For a fixed-scope, 90-day AI shop, the accessible slices are the same ones the pillar names — document automation, RAG over VA policy, internal tooling, AI features in veteran-facing products — and the Tech Sprint is the lowest-cost way to earn a credential against them. The broader build-vs-team picture is in the VA pillar.

What are the procurement mechanics that turn a pilot into a contract?

Five paths convert a successful demo into funded work, and a small SDVOSB should know all of them.

First, COMPETES Act sole-source — the Abridge/Knowtex template, where the challenge is the competition and the follow-on firm-fixed-price contract needs no separate source selection. Second, Veterans First sole-source: under 38 U.S.C. § 8127, a VA contracting officer (CO) can sole-source to a verified SDVOSB up to $5 million for services, and a Tech Sprint demo is excellent past-performance evidence to support it — the mechanics are in the Veterans First guide. Third, similarly-situated subcontracting: subbing to an SDVOSB prime like Rise8, where your work counts toward the prime’s self-performance floor under FAR 52.219-14, detailed in the teaming guide. Fourth, task orders on existing IDIQs — once a pilot is proven, follow-on work flows through T4NG2, SPRUCE, VETS 2, or GSA. Fifth, the enterprise IDIQ at scale: the VA is moving toward a dedicated multi-award “Ambient Scribe Enterprise IDIQ,” which was in protest at GAO in spring 2026.

The “valley of death” between a working pilot and a production award is real, and it’s bridged by three things: being on a vehicle when the production money lands, having the FedRAMP/ATO posture to actually deploy, and having past performance the contracting officer can cite. Tech Sprint pedigree solves the third directly — which is why it’s worth more to a brand-new firm than its modest prize money suggests.

How does a brand-new SDVOSB realistically get into this pipeline?

In three phases, and none of them require being a household name.

Phase one — earn the pedigree. Apply to the next Tech Sprint round through Challenge.gov. There’s no fee, you keep your IP, and you build on synthetic data, so there’s no PHI exposure. The win isn’t necessarily first place — being named a finalist is itself a durable, citable past-performance signal. For a “working software, not strategy decks” shop, the format is unusually well-aligned: the sprint rewards a working prototype on a 90-to-120-day clock, which is the same cadence as a 90-day fixed-scope build.

Phase two — convert it. File a SAM.gov capability statement that names the Tech Sprint round and problem statement; monitor SAM.gov for sources-sought notices in that problem space and respond fast; pitch VA program offices (NAII, OCTO, the VHA Digital Health Office) a fixed-price proposal under the $5M Veterans First ceiling; and approach the SDVOSB primes already winning AI-scaling work — Rise8 is the obvious target — as a similarly-situated subcontractor. Phase three — sustain: get onto T4NG2 or SPRUCE as a sub or teaming partner, pursue task orders, and re-enter the next sprint to refresh the credential.

A timing note worth acting on: the VA AI program is in a leadership transition — its long-serving CTO and Chief AI Officer announced his departure in March 2026, with acting leaders now in place — and the FY27 budget request keeps a dedicated automation-and-AI line. Transitions plus fresh budget lines are exactly when a new vendor can register on the radar.

Does any of this require CMMC?

No — and stating it correctly to a contracting officer is part of being credible.

VA AI pilots run inside the VA Enterprise Cloud (VAEC) on AWS GovCloud and Azure Government, both at FedRAMP High, governed by the VA ATO process, VA Handbook 6500, NIST 800-53, and HIPAA where clinical data is involved, with Section 508 for anything user-facing. The sprint phase itself uses synthetic or de-identified data, so there’s no production-PHI exposure while you’re proving the concept. CMMC — the Cybersecurity Maturity Model Certification — is a Department of Defense framework for protecting controlled unclassified information on DoD contracts; it is not a VA requirement. For a Cloudflare-native shop, the honest and complete posture is FedRAMP-aware, designed to inherit FedRAMP High controls from the VA Enterprise Cloud — not CMMC-certified, and CMMC isn’t required for VA AI work. The full framing is in FedRAMP-aware, not CMMC-certified.

Working with Truvisory

Truvisory is an SBA-verified SDVOSB founded by a combat veteran, building working AI and automation on a Cloudflare-native, FedRAMP-aware architecture — fixed-scope, in 90 days, on the same clock a Tech Sprint runs on.

If you’re a VA program office sponsoring an AI pilot that needs a production path, or an SDVOSB prime like Rise8 scaling AI work and needing similarly-situated depth, start with a scoping call. For the full VA picture, see the VA AI modernization pillar; for the vehicles this work lands on, see the T4NG2 and SPRUCE guides.