Mapping your AI/ML work to the Dhanasar framework
USCIS evaluates every NIW petition against the three-prong framework set out in Matter of Dhanasar. Every exhibit and letter you assemble should trace back to one of these three prongs.
Prong 1: Substantial merit and national importance
The endeavor itself — not your resume — must have substantial merit (technical, economic, or otherwise) and national importance. This is evaluated on the endeavor's potential effects, not your personal credentials.
Prong 2: Well positioned to advance the endeavor
This asks whether you, specifically, have the training, track record, and resources to carry the endeavor forward — evidenced by publications, past projects, github repositories, industry partnerships, or funding already secured.
Prong 3: Benefit to the US of waiving the job offer
Here you address why requiring a labor certification and a specific employer sponsor would be impractical or counterproductive given the endeavor's nature — for example, work that spans multiple institutions or moves faster than a single job posting can accommodate.
Worked contrast
| Endeavor statement | Prong 1 | Prong 2 | Prong 3 |
|---|---|---|---|
Defining your proposed endeavor in concrete, evaluable terms
An endeavor statement needs enough specificity that an officer could, in principle, check whether you did what you said. Build it from four components.
Step 1: Name the technical problem
State the gap or limitation your work addresses, not the general field. "Improving AI safety" is not a technical problem; "reducing false-negative rates in transformer-based intrusion detection for industrial control systems" is.
Step 2: Specify the methodology
Name the actual approach — architecture, training regime, dataset type, evaluation protocol. This ties the statement to verifiable publications, code, or deployments.
Step 3: Identify the population or industry affected
State who uses or is affected by the outcome: a sector (power utilities, radiology departments), a user base, or a set of downstream systems.
Step 4: Add quantifiable scope
Where available, cite numbers: users served, systems deployed, funding secured, publications citing the method, or organizations adopting it.
Before: "I work on advancing machine learning for important real-world applications." After: "I develop interpretable anomaly-detection models for power-grid security, deployed in utility monitoring pipelines covering multiple regional grid operators, with methodology published in two peer-reviewed venues and adopted in one open-source toolkit with measurable download volume."
Checklist
- Too broad: no named technical problem, no methodology, no affected population, no numbers.
- Specific enough: a reader unfamiliar with you could verify each claim against a publication, repository, or letter.
- Avoid restating a job title or employer description — describe the endeavor, not the position.
Evidence categories specific to AI/ML careers
USCIS evaluates evidence by what it demonstrates, not by its label. For each item below, document not just that it exists but what it shows about the endeavor's merit, reach, or your positioning.
Core evidence types
- Peer-reviewed publications. Papers at venues such as NeurIPS, ICML, ICLR, CVPR, ACL, or in journals (e.g., JMLR, IEEE Transactions series) establish technical contribution. Note acceptance rates and venue tier where available.
- Citation counts. Pull from Google Scholar and Semantic Scholar. Raw counts mean little alone — pair them with field-normalized context (median citations for papers of similar age/subfield) so an officer can judge significance.
- Open-source metrics. GitHub stars, forks, and contributor counts; PyPI or Hugging Face download counts. These carry weight when tied to adoption by identifiable organizations or downstream projects, not as standalone numbers.
- Patents. Issued or published applications tied to the endeavor's technical claims.
- Program committee/reviewer roles. Service at the venues above signals recognized expertise; list specific conferences and years.
- Benchmark leaderboard placements. Rankings on Papers with Code or similar trackers, with the specific task, dataset, and metric named.
- Industry deployment letters. Statements from organizations describing actual use of the work, ideally with a measurable outcome (cost reduction, accuracy improvement, incidents prevented).
Across all categories, evidence lacking a tied, measurable outcome reads as activity rather than impact.
Building the citation and impact index
Metrics scattered across profiles are hard for an officer to verify and easy for you to misstate. Build a single table before drafting any evidence exhibit.
Step 1: Pull the raw data
- Google Scholar: open your public profile, record total citations, h-index, and per-paper citation counts.
- Semantic Scholar: use the author API or the public profile page to cross-check citation counts and get per-paper influential-citation figures, which Google Scholar does not provide.
- GitHub: use the repository's Insights tab for stars, forks, and traffic; use PyPI or Hugging Face download statistics pages for package or model downloads.
Step 2: Organize into one table
Use a spreadsheet with these columns: Work/Repo, Venue/Platform, Date, Metric type, Metric value, Source URL, Date retrieved. One row per metric, not per work — a single paper may have separate rows for Google Scholar citations and Semantic Scholar citations, each dated.
Step 3: Archive each source
For every row, save a PDF or screenshot of the page showing the number, with the URL and date visible. Store these in a folder matching your exhibit numbering so each figure in the petition letter can be traced to a saved record.
Step 4: Refresh close to filing
Metrics change continuously. Re-pull every figure within a few weeks of filing, update the Date retrieved column, and re-archive screenshots so the evidence reflects current standing rather than numbers gathered months earlier.
Recommendation letters: selecting and briefing independent experts
Choosing writers who count as independent
An officer reading a letter from a current supervisor, PhD advisor, or frequent coauthor weighs it as an interested party's view. Build a letter roster weighted toward people who have no coauthorship, no employment relationship, and no financial stake in the petitioner's work: researchers at other institutions who cite or build on the work, engineers at companies that deployed a model or library, program chairs or area chairs who handled the petitioner's submissions, or maintainers of downstream projects. A mix of two or three independent letters alongside one or two from close collaborators is more useful than a stack of letters from the petitioner's own lab or team.
What each letter should cover
Each writer should address: how they know the work (reading a paper, using a tool, reviewing a submission, replicating a benchmark result), the specific technical contribution they can speak to, and an independent observation of impact — adoption inside their own organization, a measurable improvement they obtained using the petitioner's method, or a citation/comparison in their own published work.
Briefing without drafting their words
Send each writer a short outline, not a draft: (1) relationship and how they encountered the work, (2) one or two specific contributions, (3) one concrete, measurable outcome they observed, (4) closing statement in their own words. Supply dates, paper titles, and metrics for accuracy, but let the writer compose the letter.
Organizing the Form I-140 NIW petition package
USCIS adjudicators work through a petition in the order it is bound. A package that follows a predictable order, with everything labeled and indexed, is easier to read and easier to check against for missing pieces.
Filing order
- Form I-140. Confirm the current edition and any required supplements on uscis.gov before filing; using an outdated edition can cause a rejection.
- Filing fee payment. Confirm the current I-140 fee and payment instructions on uscis.gov.
- Asylum Program Fee payment. This is a separate payment from the I-140 filing fee, made by the same payment method (check or money order, or Form G-1450 for a card). Confirm the current amount and instructions on uscis.gov.
- Petition letter (the cover letter arguing the three Dhanasar prongs).
- Endeavor statement, if filed as a standalone document rather than embedded in the petition letter.
- Table of contents for exhibits.
- Evidence exhibits, indexed as Exhibit A, Exhibit B, Exhibit C, and so on, each with a cover sheet naming its contents.
- Recommendation letters, usually grouped as a labeled exhibit range.
- CV, placed as its own exhibit at the end.
File naming
For electronic assembly before printing or e-filing, name files so the exhibit letter and content are visible without opening the file: ExhibitA_Publication_NeurIPS2023.pdf, ExhibitB_GitHub_Metrics_ModelRepo.pdf, ExhibitC_Letter_DrSmith.pdf. Keep the same labels in the petition letter's internal citations so a reader can move between the narrative and the exhibit without searching.
Filing logistics: premium processing and tracking your priority date
Premium processing
Form I-140 petitions in the EB-2 NIW classification are eligible for premium processing. Timeframes are counted in business days, not calendar days. To request it, file Form I-907 together with (or after) the I-140, following the current instructions on uscis.gov. Fees and processing timeframes change periodically — confirm both on the USCIS Fee Calculator and the Form I-907 page before paying, and do not rely on a figure from an old filing.
Locating your chart in the Visa Bulletin
Your priority date is the receipt date USCIS assigns your I-140. To see when it becomes current, open the current Visa Bulletin on travel.state.gov and find the table for your category (EB-2) and your country of chargeability. Two charts appear each month: Final Action Dates and Dates for Filing. USCIS publishes, usually on its visa bulletin page, which chart applicants should use for filing Form I-485 or for consular processing that month — check this each time, since it can change.
Comparing dates
Find the cell at the intersection of your category and country. That date is the cutoff. Your priority date is current only if it is earlier than the cutoff date shown, or if the cell shows "C." A priority date equal to the cutoff is not current. A cell showing "U" means unavailable. Cutoff dates can advance from one month to the next or retrogress to an earlier date or to "U." Recheck the bulletin and uscis.gov monthly rather than assuming a prior month's figures still apply.
Common errors specific to AI/ML petitions
Citation counts without context
A raw citation number means little to an officer without a frame of reference. Listing "312 citations" without noting the paper's age, venue norms, or how many co-authors share credit invites the reviewer to guess. Pair every citation figure with field context: typical citation velocity for the subfield, the paper's position relative to others from the same venue and year, and the petitioner's specific contribution if the paper has many co-authors.
Letters that restate the CV
A letter that lists degrees, job titles, and publication counts the officer can already see in the CV adds nothing. It reads as padding and can prompt a request for more probative evidence. Each letter should describe specific, independently observed facts about the work's effect — not a paraphrase of the petitioner's own materials.
Endeavor statements that describe a job function
Statements like "building machine learning systems for my employer" describe a role, not a specific endeavor, and give the officer nothing concrete to evaluate against the three prongs.
Outdated metrics
GitHub stars, download counts, and leaderboard rankings change constantly. A figure pulled months before filing may no longer match what the officer finds if they check independently. Refresh these close to filing and note the retrieval date on each exhibit.
Disorganized exhibits
Unlabeled attachments, missing cross-references between the petition letter and the exhibit list, or exhibits out of sequence force the officer to hunt for support, which is a frequent driver of RFEs.
You may also like
EB-2 NIW for Colombian and Venezuelan Professionals
Learn how to assemble foreign credential, employment, and country-condition evidence from Colombia and Venezuela into a self-filed EB-2 NIW petition.
Read →EB-2 NIW for Chinese Researchers and Professionals
Learn what extra documentation Chinese-trained EB-2 NIW applicants need — degree verification, CHESICC/学信网 reports, employment records, and certified translations — and how to get it accepted.
Read →EB-2 NIW for Indian IT and Engineering Professionals
Learn how to map your IT or engineering career onto the three Dhanasar prongs and assemble a petition file that accounts for the country-specific backlog realities Indian applicants face.
Read →