EB-2 National Interest Waiver Matter of Dhanasar Analysis

Building AI systems that widen the aperture of American education for the learners its laws already name: the underserved, & the neurodiverse.

Petitioner Uchenna W. Nwoke

Category EB-2 · Advanced Degree /
Exceptional Ability

Framework Matter of Dhanasar,
26 I&N Dec. 884 (AAO 2016)

Statute 8 U.S.C.
§ 1153(b)(2)(B)(i)

Current role Software Development Engineer,
Amazon (since 2024-10-21)

Prepared 2026 · under
attorney review

Vision & mission

An endeavor of substantial merit and national importance under the Dhanasar framework.

The proposed endeavor is to design, ship, and maintain AI-powered software that lowers the friction between educational content and the U.S. learners least served by existing tools — students in under-resourced school districts, English-language learners, and neurodiverse adults, particularly those on the autism spectrum.

The work runs along three overlapping surfaces: assistive reading and comprehension tools; predictive and screening support informed by the empirical study of adult autism spectrum disorder; and contributions to the open-source infrastructure on which trustworthy AI systems will be built and audited. Together these are not a portfolio of unrelated projects but a single line of work: making the accommodations that federal law already promises — meaningful access, individualized support, appropriate evaluation — deliverable at software cost per learner.

Prong I

Substantial merit and national importance

Prong II

ii.

Well-positioned to advance the endeavor

Prong III

iii.

On balance, beneficial to waive labor certification

§ 1.1Substantial merit

The endeavor addresses statutory commitments the United States has already made — under the Individuals with Disabilities Education Act, Section 504 of the Rehabilitation Act, and Title II of the Americans with Disabilities Act — to provide meaningful access to education for learners with disabilities and for those historically underserved. AI systems that reduce the marginal cost of delivering that access are not merely useful; they are on the critical path between statutory promise and lived outcome.

§ 1.2National importance

Educational under-attainment among the target populations has demonstrated downstream effects on labor-force participation, healthcare utilization, and long-run federal expenditure. The endeavor's benefits are not localized to any single district, employer, or clinical setting — the tooling, research, and infrastructure are portable across state lines and educational contexts, which is the standard the AAO has recognized as satisfying national importance in Dhanasar and its progeny.

§ 1.3Positioning

The petitioner's positioning combines applied engineering at a hyperscale employer, published and in-progress ML research on adult ASD, verified open-source contributions to the infrastructure layer on which trustworthy deployment depends, and prior direct work with organizations serving underrepresented learners. Recommender letters address each of these surfaces from persons with first-hand knowledge of the work.

II.

About the petitioner

An engineer working at the intersection of applied ML, distributed systems, and access.

Uchenna Nwoke is a Software Development Engineer at Amazon, where he has worked since October 21, 2024. His work sits at the intersection of applied machine learning, distributed systems, and human-centered software design.

§ 2.1Employment

At Amazon, the petitioner ships software at scale within a large engineering organization, with concrete production ownership and measurable business impact. Prior to Amazon, he held engineering roles that placed him proximate to educational and healthcare-adjacent problem spaces, informing the endeavor's framing.

§ 2.2Research direction

The petitioner's research direction concerns predictive and characterization modeling of adult autism spectrum disorder — an area where the peer-reviewed base has grown substantially but where deployable, statistically defensible screening tools remain sparse. This work is informed by taxonomies of bias in human-computer and human-robot interaction and by fairness-aware ML methodology.

§ 2.3Open-source practice

The petitioner contributes to public infrastructure projects that materially affect the substrate on which trustworthy AI systems will be deployed. Verified merged contributions to Reth, the Rust Ethereum execution client stewarded by Paradigm, are documented in Section V.

§ 2.4Access-oriented work

Alongside the above, the petitioner has contributed to organizations focused on widening access to educational opportunity for under-resourced and underrepresented learners — the population whose outcomes the proposed endeavor is designed to move.

Epigraph · 20 U.S.C. § 1400(d)(1)(A)

“To ensure that all children with disabilities have available to them a free appropriate public education that emphasizes special education and related services designed to meet their unique needs.”

Individuals with Disabilities Education Act
Purpose clause · codified 1990, most recently amended 2004

III.

The gap · the response

Where existing systems fall short, and what an AI-native response looks like.

The accommodations the United States is committed to providing are, in practice, gated by cost and expertise — two constraints AI systems can materially relax if built with the right priorities.

A.The gap

~7.5M^[verify]

U.S. children ages 3–21 receive special-education services under IDEA — roughly 15% of public-school enrollment.

NCES, Digest of Education Statistics [verify vintage]

31%^[verify]

Share of U.S. 4th-graders performing at or above NAEP "Proficient" in reading — persistently lower for students eligible for federally funded programs.

NAEP Reading Assessment [verify vintage]

Years^[verify]

Adult autism spectrum disorder is diagnosed at a median delay of years relative to childhood diagnosis, with women and racial minorities disproportionately underdiagnosed.

Peer-reviewed literature synthesis [verify citations]

Uneven

Access to trained evaluators, assistive technology, and one-on-one instructional support varies sharply by ZIP code, income, and race — the constraints AI systems are structurally positioned to relax.

OSEP Annual Report to Congress; state-level LEA data

B.The response

Multi-modal document understanding

Closing the distance between text-heavy educational materials and learners who need audio, structural, or plain-language variants — at software cost per learner rather than per human hour.

ii.

Adaptive comprehension support

Interfaces that adjust pace, terminology, and modality based on learner interaction — not only correctness — grounded in the same document the learner is trying to access.

iii.

Screening support for adult ASD

Statistically informed triage tools that flag likely-eligible individuals for professional evaluation, with explicit fairness auditing — supporting rather than displacing clinician judgment.

iv.

Verifiable trust substrate

The underlying infrastructure — provenance of models, integrity of deployment, auditability of outputs — must be built into the systems, not layered on afterward. This is where the open-source contributions in Section V connect back to the endeavor.

Practitioner-facing tooling

Reducing the paperwork and coordination burden on the teachers, IEP teams, and clinicians who deliver the human work — so their time compounds instead of being consumed by process.

IV.

Evidence of prior work

Three concrete surfaces of prior and ongoing work that inform the endeavor.

Project 01

Chat with PDF

AI reading & comprehension tool · Author

A document reader that lets users ask natural-language questions of PDF documents and receive answers with document-anchored citations, without conventional keyword search. Particularly useful for learners for whom skimming long texts is inaccessible — whether due to reading disability, cognitive load, or unfamiliarity with domain vocabulary.

The tool sits at the intersection of retrieval-augmented generation and accessibility engineering: outputs are grounded in the source document rather than free-generated, which is the correctness posture educational and clinical settings require.

SurfaceLearner-facing

Endeavor linkMulti-modal document access

StatusAdoption metrics pendingverify

EvidenceScreenshots, code, deployment

Project 02

Modeling Autism in Adults

Applied ML research · Contributor

Applied machine-learning work on the prediction and characterization of adult autism spectrum disorder. The work spans dataset curation, model architecture selection, and — importantly — fairness auditing informed by taxonomies of bias in human-computer and human-robot interaction.

The research direction is not "detect autism" as a marketing claim but the narrower, more defensible target of building triage support that improves the throughput of professional evaluation, particularly for the demographic subgroups the peer-reviewed literature identifies as underdiagnosed.

SurfaceResearch · clinical support

Endeavor linkNeurodiverse-learner outcomes

CollaboratorDr. Lisa Milkowskiverify

DeliverableTaxonomy of Bias in HRIverify

Project 03

AVELA

Educational access & advocacy · Contributor

Work with AVELA, an organization focused on widening access to education for historically underrepresented learners. The contribution translates access-oriented pedagogy into deployable, low-friction software — the layer where research infrastructure meets classroom reality.

The AVELA work grounds the endeavor's population target in something more concrete than statistics: direct exposure to the coordination frictions, resource gaps, and instructional-time constraints that any deployable tool will need to respect.

SurfaceCommunity & access

Endeavor linkUnderserved-learner reach

RoleTitle pendingverify

EvidenceReference letter, outreach records

Public infrastructure contributions

Verified contributions to the substrate on which trustworthy AI deployment depends.

The endeavor is not only about the surface tools learners see. The infrastructure layer — how software is built, distributed, and verified — determines whether accessibility work can be trusted at scale.

Reth Paradigm · Rust

A high-performance Ethereum execution client written in Rust. Widely used across the Ethereum research, validator, and infrastructure community. Contributions to Reth touch the correctness, performance, and observability of a system that runs adversarial workloads at scale — engineering discipline that transfers directly to safety-critical AI infrastructure.

3 merged PRs verified paradigmxyz/reth

Calimero Distributed compute

Contributions to Calimero's distributed application infrastructure. The relevance to the endeavor is architectural: how decentralized, provenance-preserving compute can carry AI workloads to educational and clinical settings without concentrating trust in a single operator.

Contributions documented calimero.network

§ 5.1Verified merged pull requests

The following contributions have been confirmed via direct GitHub API verification and are cited in the petition record. Additional pull requests referenced in earlier drafts have been withdrawn where the API check showed they were closed without merging — the record is limited to what can be defended.

PR	Status	Repository	Contribution
№18440	Merged	paradigmxyz/reth	Description drawn from actual diff content only. draft
№18558	Merged	paradigmxyz/reth	Description drawn from actual diff content only. draft
№18661	Merged	paradigmxyz/reth	Description drawn from actual diff content only. draft

Evidentiary accuracy is non-negotiable. Petition language must match verifiable facts; every PR description here will be rewritten from the actual diff before filing, not from inflated summaries. — Internal drafting principle

VI.

Proposed platform

A high-level architecture for the systems the endeavor is set up to build.

The architecture organizes the endeavor's technical surfaces as four layers: who the platform serves, what they interact with, what powers those interactions, and what makes the whole assembly trustworthy.

Figure I · caption

Layers 1–3 describe the platform surface; Layer 4 describes the substrate that makes the assembly trustworthy. The trust layer is not decorative — it maps onto the open-source infrastructure work described in Section V and is the reason those contributions are cited alongside the learner-facing work rather than as a separate portfolio pillar.