The EB-2 NIW industrial data scientist had reduced equipment interruptions, identified energy waste, and built analytics used by plant engineers. His public record still described an employee who developed models for one company. The NIW case became credible after two implementations were reconstructed, a transferable manufacturing analytics method was published, outside professionals used the public tools, and a lawful multi-client U.S. plan showed how the work could continue beyond one sponsoring employer.

This is an anonymized composite case study drawn from completed professional profile development and EB-2 national interest waiver petition work. It does not identify a specific client. Names, employers, plants, dates, equipment, datasets, model details, publication titles, user organizations, performance values, letters, and selected implementation facts have been combined, withheld, rounded, or adjusted to protect privacy, trade secrets, cybersecurity, and confidential business information.

Case at a glance

Case element	Completed case record
Profession	Industrial data science, predictive maintenance, manufacturing analytics, condition monitoring, energy performance analysis, and operational decision support
Starting point	A master’s trained data scientist in H-1B status with approximately eight years of combined manufacturing and analytics experience, strong employer results, two coauthored papers, and limited independent recognition
Expert specialization	Predictive maintenance and production normalized energy optimization for small and midsize U.S. manufacturers
Main profile problem	The record showed valuable employment but did not identify the client’s personal analytical decisions, separate reusable methods from employer owned code, or establish an endeavor that could continue beyond one H-1B employer
Profile-building period	Approximately fourteen months before filing, followed by petition preparation and adjudication
What already existed	Sensor histories, vibration and temperature data, maintenance work orders, failure records, production logs, energy meter data, model notebooks, pilot reports, internal presentations, and managers able to confirm the client’s role
What Advance My Profile organized or developed	A contribution chronology, the Manufacturing Reliability and Energy Analytics Method, two first author technical works, a synthetic public benchmark, an implementation workbook, external training, completed peer evaluation, independent use records, letters of future interest, and a multi-client U.S. professional plan
NIW evidence emphasized	The broader implications of repeatable reliability and energy analytics for manufacturing; the client’s completed implementations, authorship, public assets, outside use, teaching, and peer evaluation; and the benefit of allowing the endeavor to proceed across organizations rather than through one permanent job offer
What was deliberately not pursued	Employer code or raw data, unsupported patents, open memberships, paid publicity, internal model reviews described as judging, savings figures that could not be normalized, claims that analytics prevented every failure, and future consulting performed without appropriate work authorization
Petition result	USCIS approved the Form I-140 EB-2 NIW petition without issuing a Request for Evidence
Procedural limit	The I-140 approval established the immigrant petition classification only. It did not itself grant permanent residence, change H-1B employment authorization, permit outside work, or authorize the client to begin the proposed independent engagements.

The H-1B record proved employer value, not professional independence

At intake, the client had a strong employment record. He had built condition monitoring models, supported plant pilots, analyzed equipment failures, and helped engineering teams investigate abnormal energy use. Performance reviews described him as reliable and technically capable. Internal dashboards showed measurable improvement. None of that, by itself, explained an employer independent proposed endeavor.

The resume was organized around tools and job duties: Python, SQL, cloud pipelines, anomaly detection, forecasting, dashboards, and stakeholder meetings. The project descriptions named plants and model accuracy but did not show which analytical choices belonged to the client, how plant personnel used the outputs, or whether the work could be adapted outside the company. The same problem appeared in the recommendation letters. Supervisors praised performance but described the client mainly as a valued employee.

The immigration constraint was also misunderstood. A national interest waiver can remove the job offer and labor certification requirements for the immigrant petition, but approval does not erase the conditions of H-1B employment. The development plan therefore separated future NIW independence from current work authorization. The client completed profile-building activities that were lawful under his circumstances and did not begin outside consulting merely to create evidence.

Legal context: USCIS evaluates whether the petitioner qualifies for EB-2 and whether the proposed endeavor has substantial merit and national importance, the person is well positioned to advance it, and it would benefit the United States to waive the job offer and labor certification requirements. The analysis concerns the specific endeavor, not the general importance of artificial intelligence, manufacturing, or energy efficiency.

The audit separated ordinary data science work from attributable manufacturing judgment

Data scientists routinely clean data, train models, compare metrics, build dashboards, and present findings. Those tasks were not described as original contributions merely because the client performed them well. The audit looked for decisions that changed how a plant defined a problem, collected evidence, evaluated uncertainty, triggered action, or verified an operational result.

We reconstructed each project from contemporaneous records rather than drafting broad letters first. The source archive included issue tickets, model version history, sensor quality notes, maintenance records, plant meeting minutes, production normalization worksheets, validation results, and release approvals. The chronology identified the problem before the client’s involvement, the options considered, the client’s analytical decision, the action taken by plant personnel, and the later outcome.

Four decisions emerged repeatedly. The client separated operating states before modeling, rejected labels that could not be tied to a verified event, used uncertainty and abstention rules instead of forcing every prediction, and connected each alert to an owner and a permitted maintenance or energy review action. These were not generic software choices. They reflected his understanding of how industrial data is produced and how plant teams make decisions under cost, safety, and production constraints.

Confidentiality remained a real limit. The employer would not release raw sensor streams, production recipes, asset identifiers, source code, customer names, plant network diagrams, or commercial savings calculations. We used redacted version histories, approved screenshots, aggregate results, code contribution summaries, data custodian letters, and witness statements. One apparently strong project was removed because a major equipment replacement occurred during the comparison period and the available records could not isolate the effect of the client’s model.

The endeavor was narrowed to manufacturers that could use the method

The first proposed endeavor was to use artificial intelligence to modernize U.S. manufacturing. It covered maintenance, energy, quality, supply chains, robotics, worker safety, and cybersecurity. The description was too broad to show what the client would actually deliver, which organizations could use it, or how the prior record supported future execution.

The final endeavor focused on developing and implementing practical predictive maintenance and production normalized energy optimization systems for small and midsize U.S. manufacturers. Intended users included component manufacturers, food and beverage plants, packaging operations, light chemical processors, industrial service companies, and other facilities that had useful operating data but limited internal data science capacity.

The endeavor did not promise autonomous maintenance, universal failure prediction, or a fixed percentage of energy savings. It did not replace reliability engineers, maintenance technicians, plant managers, safety reviews, or process engineers. The client’s work concerned data readiness, operating state analysis, model validation, alert design, decision support, outcome verification, and transfer to plant personnel.

Manufacturing context: NIST describes predictive maintenance as maintenance initiated from failure predictions using observed condition data such as temperature, sound, and vibration. NIST also studies prognostics and health management for smart manufacturing. The Department of Energy’s Better Plants program supports manufacturers seeking measurable energy, water, waste, reliability, and competitiveness improvements. These sources established field context; they did not prove the client’s individual importance.

H-1B compliance changed the order of profile development

The profile could not be built by asking the client to perform unauthorized independent services. During the development period, his paid work remained within the terms of his H-1B employment. Public writing used information he was permitted to disclose. Conference participation and peer review were handled through approved professional activities. The outside use evidence came from freely available tools and publications, not from paid consulting engagements performed without authorization.

Letters from prospective U.S. users were also framed accurately. They described future interest in pilots, training, or collaboration after the client had appropriate authorization and after the organizations completed their own contracting, data security, and operational reviews. The letters did not claim that projects had already started or that the NIW petition itself would authorize the work.

This distinction strengthened the case. It showed that the proposed endeavor was independent of one permanent job offer while the client continued to comply with the temporary status under which he was working at the time of filing.

The Manufacturing Reliability and Energy Analytics Method made the work transferable

We organized the completed work into a seven stage method. The name described the client’s own sequence for turning plant data into controlled decisions. It was not presented as a new statistical theory, an industry standard, or a replacement for established maintenance and energy management practices.

Stage	What the client developed	Evidence preserved
1. Rights, system, and decision boundary	Defined the plant process, assets, users, data ownership, cybersecurity limits, prohibited uses, and the operational decision the analysis was allowed to support.	Data rights notes, system maps, access approvals, decision records, security reviews, and scope statements.
2. Data and operating state map	Mapped sensors, maintenance records, production states, product mix, downtime, energy meters, missing periods, time alignment, and known changes to equipment or controls.	Data dictionaries, tag lists, time alignment checks, production schedules, asset histories, and data quality logs.
3. Event and baseline definition	Defined verified failures, precursor windows, normal operating states, production normalized energy baselines, exclusions, and rules for uncertain labels.	Failure adjudication records, baseline worksheets, normalization notes, exclusion logs, and reviewer confirmation.
4. Model and uncertainty testing	Compared interpretable statistical and machine learning approaches across time periods, assets, operating states, class imbalance, drift, and missing data conditions.	Model cards, test matrices, threshold studies, error analyses, calibration plots, and version histories.
5. Human decision workflow	Linked each alert or opportunity to an owner, priority, permitted action, supporting evidence, escalation rule, and route for rejecting or correcting the output.	Alert taxonomies, decision matrices, work order links, feedback forms, meeting records, and training materials.
6. Pilot and outcome verification	Tested the system in a bounded setting and compared downtime, work order behavior, alert quality, energy intensity, and operational constraints against an appropriate baseline.	Pilot plans, acceptance criteria, aggregate results, production normalized comparisons, and custodian letters.
7. Transfer and monitoring	Converted the method into reusable guidance, monitored drift and later outcomes, recorded local changes, and preserved a clean evidence trail for another site.	Implementation workbook, public benchmark, training records, change logs, monitoring summaries, and independent use letters.

Contribution one: the maintenance model learned when not to issue an alert

The first contribution involved motors, pumps, and compressors used across several production lines. The employer had historical vibration, temperature, current, work order, and alarm data, but the records were not ready for predictive maintenance. Sensors had been replaced, sampling rates differed, and failure codes mixed verified mechanical events with inspections, nuisance alarms, and unrelated shutdowns.

The client created an event adjudication process with reliability engineers. He separated verified bearing, alignment, lubrication, overheating, and process induced events from weak labels. He then segmented data by operating state, excluded periods affected by planned shutdowns or sensor faults, and tested whether model thresholds remained stable across asset classes and production conditions.

The original system generated too many alerts during start-up, cleaning, and low load operation. The client added operating state gates, calibrated the anomaly score, and introduced an uncertain category that withheld an alert when the evidence was incomplete. Each released alert identified the contributing signals, the relevant asset history, and the maintenance action that plant personnel could consider. The maintenance team retained authority to inspect, defer, or reject the recommendation.

Measure	Earlier record	Later record	Evidence boundary
Alerts confirmed as actionable after engineering review	Approximately 49 percent	Approximately 78 percent after operating state and uncertainty controls	Measured in the defined pilot assets; it did not establish performance for every asset or facility.
Emergency work orders among maintenance actions on the pilot assets	About 36 percent	About 21 percent over the later comparison period	The comparison disclosed production changes and excluded planned capital replacements.
Median warning time for verified degradation events	Approximately 2.4 days	Approximately 7.1 days for the events meeting the final evidence rules	Applied only to the verified event set and did not mean every failure was predictable.
Unplanned interruption hours per 1,000 operating hours	Approximately 7.6 hours	Approximately 4.5 hours	Showed an associated operational improvement; maintenance changes occurred during the same period.

The petition did not claim that the model prevented every breakdown or caused the entire reduction in interruption time. It showed that the client corrected the labeling and operating state problem, improved alert quality, created a usable maintenance workflow, and verified implementation with records that plant personnel understood.

Contribution two: energy analytics became useful only after production was normalized

The second contribution involved a manufacturer whose monthly energy reports showed wide variation but did not explain whether the change came from production volume, product mix, shifts, weather, downtime, cleaning, or equipment operation. A simple trend line identified high use months but produced weak recommendations.

The client built a production state model that separated active production, changeover, cleaning, idle, start up, and shutdown periods. He linked electricity and fuel use to production counts, run time, product family, and selected environmental conditions. He also created rules for excluding missing or unreliable meter intervals and for flagging changes that required plant investigation rather than automatic action.

The analysis identified two recurring sources of avoidable use: equipment remaining in a high demand state during extended idle periods and unstable utility demand during product changeovers. Plant engineers reviewed the findings, adjusted schedules and control sequences, and chose which recommendations could be implemented safely. The client then compared energy intensity using the agreed normalization method instead of selecting one favorable utility bill.

Measure	Earlier record	Later record	Evidence boundary
Electricity intensity for the covered production area	Baseline indexed at 100	Approximately 91.2 over the later normalized period	Adjusted for production volume and selected product mix differences; not a whole plant claim.
Off-shift electrical demand during qualifying idle periods	Baseline indexed at 100	Approximately 82 after schedule and control changes	Applied only to intervals meeting the defined idle state rules.
Energy opportunity flags confirmed by plant review	Approximately 44 percent	Approximately 73 percent after state segmentation and revised thresholds	Showed improved review value; confirmation did not guarantee that every action was economical.
Changeover periods exceeding the agreed energy intensity threshold	About 31 percent	About 17 percent in the later comparison period	Other operational improvements occurred during the same period and were disclosed.

The contribution was not a claim that the client invented industrial energy management or controlled the plant. His work made the comparison fairer, connected abnormal use to operating conditions, and gave plant engineers a repeatable way to investigate and verify changes.

A synthetic benchmark solved the public evidence problem

The strongest employer datasets could not be published. Releasing raw records would have exposed equipment identity, production schedules, customer information, plant vulnerabilities, and proprietary operating practices. A generic article without data would have been easier to publish but less useful to other professionals.

The client created a synthetic benchmark that reproduced the analytical problems without copying the employer’s values. It contained multiple operating states, sensor drift, missing intervals, rare degradation events, maintenance labels of different quality, and production normalized energy records. A technical note explained how the data were generated, which relationships were artificial, and why the benchmark could not be treated as evidence of a real plant’s performance.

The public package also included a model evaluation worksheet, alert review template, energy normalization checklist, and example documentation for uncertainty and rejected outputs. Employer counsel and a technical manager reviewed the materials before release. The final asset contained no employer code, customer names, plant configurations, or confidential thresholds.

The publication program grew from completed work, not a paper count target

The client already had two coauthored academic papers, but neither established his current specialization. We did not plan a series of unrelated manuscripts merely to add publications. The authorship program was limited to subjects supported by his completed implementations and disclosure rights.

The first first author methods article examined why industrial predictive maintenance projects fail before model selection. It addressed event labels, operating state segmentation, false alerts, uncertainty, and maintenance decision design. The article used the synthetic benchmark and disclosed that the examples were instructional rather than customer results.

The second work addressed production normalized energy analytics for facilities with sparse metering and changing product mix. It explained baselines, exclusions, state definitions, verification periods, and the difference between an analytical opportunity and an implemented engineering change. A manufacturing systems conference accepted the paper after peer review, and the client presented the work in a technical session.

Later citation and use evidence was reviewed by context. The petition did not rely on a raw citation count. It identified independent authors who used the benchmark, discussed the event labeling method, or compared their own maintenance workflow with the client’s published approach.

Outside use showed that the work had moved beyond one employer

The public benchmark and workbook produced the most useful independent evidence. A university manufacturing laboratory used the dataset in a graduate reliability course. An industrial analytics consultancy adapted the alert review template for an internal demonstration. A maintenance software provider used the operating state examples during a product validation exercise. None of these organizations employed or supervised the client.

The evidence archive preserved download records, repository history, course material, correspondence, adaptation notes, and letters explaining how the users applied the materials. The letters did not simply call the client an expert. They described the specific asset used, the reason it was selected, the changes made locally, and the professional value of the method.

A proposed claim based on anonymous repository traffic was excluded. Download counts could not prove who used the files or whether the downloads reflected meaningful professional reliance. The final record relied on identifiable use that could be explained and verified.

Teaching and peer evaluation followed visible technical work

After the methods article and benchmark were public, the client delivered a manufacturing analytics webinar and a technical workshop for reliability and energy professionals. The sessions used synthetic examples and focused on data rights, event labels, operating states, model uncertainty, alert ownership, and outcome verification. Attendance records, agendas, slides, questions, and organizer letters were preserved.

The public record later led to invitations to review conference submissions and journal manuscripts concerning industrial analytics, prognostics, condition monitoring, and manufacturing energy data. Only completed reviews were counted. Internal code reviews and employer model approval meetings remained employment evidence; they were not described as judging the work of others.

The client also participated in a technical working group concerned with manufacturing data and asset condition information. The petition described his completed comments and contribution records. Attendance alone was not presented as standards authorship or selective professional membership.

The U.S. plan showed independence without pretending to be a national platform on day one

The first draft proposed a nationwide industrial artificial intelligence platform. It required data access, integrations, financing, and adoption that the client did not yet have. The revised plan began with bounded pilots and a service model that smaller manufacturers could realistically evaluate.

Implementation phase	Completed plan design	Output identified in the plan
Phase 1: two pilot facilities	The completed plan defined one equipment reliability problem and one energy performance problem for each pilot site, subject to contracting, data rights, cybersecurity review, and appropriate work authorization.	Baseline and rights assessment, pilot model, decision workflow, verification report, and plant training.
Phase 2: transfer package	The completed plan provided for converting verified pilot lessons into plant specific documentation, a reusable training module, and revisions to the public workbook without disclosing participant data.	Implementation guide, local change record, monitoring plan, and anonymized methods update.
Phase 3: multi-site delivery	The completed plan provided for extending the method through manufacturing consultants, industrial service providers, research partners, or direct engagements with facilities that had suitable data and authority to act.	Additional pilots, partner training, independent use records, and comparison of site level implementation barriers.
Phase 4: professional dissemination	The completed plan provided for publishing bounded results where authorized and teaching the method through manufacturing, reliability, energy, and data science forums.	Practice reports, conference work, workshops, peer review, and updated public tools.

The petition did not claim signed revenue contracts or completed independent pilots that did not exist. It included letters showing reasoned future interest from manufacturers and an industrial engineering firm, together with the conditions each organization would require before work began.

The evidence addressed the three NIW questions separately

NIW element	How the completed record addressed it
Substantial merit and national importance	The endeavor concerned manufacturing reliability, energy use, operational decision quality, and transferable methods for facilities that often lack internal analytics teams. The argument focused on repeatability, multi-site use, professional dissemination, and broader manufacturing implications rather than the economic value of one employer.
Well positioned to advance the endeavor	The record combined two completed implementations, traceable personal decisions, measured results, first author publications, a public benchmark, independent use, teaching, peer evaluation, employer confirmation, technical expertise, and a staged U.S. plan.
Benefit of waiving the job offer and labor certification	The endeavor was designed to work across manufacturers, service providers, research partners, and professional forums. Tying it to one permanent position would have limited the multi-organization implementation and dissemination plan. The waiver analysis remained separate from the client’s temporary H-1B work authorization.

Several possible claims were deliberately left out

• A patent was not filed because the strongest work concerned data preparation, validation, and implementation practices rather than a defensible standalone invention.
• Open professional memberships and routine certificates were not used as evidence of distinction.
• No internal model review, dashboard approval, or employee interview was described as external judging.
• No raw customer data, source code, plant network information, production recipe, or confidential threshold was placed in the petition or public asset.
• Energy and downtime figures that could not be normalized or confirmed by a custodian were excluded.
• The petition did not claim that predictive analytics prevented every failure, replaced maintenance judgment, or caused all reported savings.
• Prospective letters were not described as contracts, completed projects, government endorsements, or proof that future work was guaranteed.
• The client did not perform unauthorized outside consulting during H-1B employment merely to create evidence for the case.
• Paid publicity, vanity awards, and publication opportunities with weak editorial controls were rejected.

The professional record changed in a traceable sequence

Stage	Documented change
Before Profile Building	Employer centered resume; confidential plant results; broad AI identity; limited authorship; no public technical asset; no outside use; proposed work tied conceptually to one employer.
Substantive reconstruction	Two contribution files established the client’s analytical decisions, plant workflow, measurable results, limits, and confirmation from people with direct knowledge.
Public professional record	First author methods work, a synthetic benchmark, an implementation workbook, a conference paper, external training, and completed peer evaluation created verifiable public facing expertise.
Independent recognition	Outside organizations used the benchmark and templates, organizers invited the client to teach, and editors trusted him to assess professional work.
Petition readiness	A bounded multi-client plan, future interest letters, H-1B compliance record, prong specific archive, and consistent evidence language connected the past work to the proposed U.S. endeavor.
Result	USCIS approved the Form I-140 EB-2 NIW petition without a Request for Evidence. The approval did not itself authorize the independent work described in the professional plan.

What this case shows about Profile Building for industrial data scientists

The client did not need to become a general artificial intelligence celebrity. He needed an attributable specialization, a clean evidence chain, and proof that the work could be understood and used outside one company. Professional Profile Development began with plant records and decision rights, not publicity.

The strongest Profile Advancement activities were also connected. The contribution chronology supported the methods article. The article supported the public benchmark. The benchmark created independent use. Independent use led to teaching and review invitations. Those activities then supported the multi-client professional plan. Each step grew from completed technical work rather than from a generic visibility campaign.

The case also shows why immigration independence and employment authorization must not be confused. NIW self-petitioning can remove the need for a permanent sponsoring employer in the immigrant petition. It does not automatically permit an H-1B professional to begin unrelated outside employment. A credible plan can be independent in purpose while remaining accurate about when and how the work may lawfully begin.

Practical lessons for manufacturing analytics professionals

• Start with the operating decision. A model is easier to evaluate when the record explains who used it, what action it informed, and what happened afterward.
• Document weak labels and rejected outputs. In industrial work, knowing when not to issue an alert can be more valuable than a higher headline accuracy score.
• Separate production change from analytical effect. Energy and downtime comparisons should disclose normalization, concurrent interventions, exclusions, and data limits.
• Protect employer and customer information. Synthetic data, blank tools, redacted version history, and custodian letters can preserve evidence without copying protected systems.
• Build outside recognition after substance is visible. Publications, teaching, review, and independent use are stronger when they grow from a real method and usable public asset.
• Do not create immigration evidence through unauthorized employment. Present future interest and implementation plans accurately.
• Organize the NIW record by the three legal questions. A long achievement list does not replace a clear endeavor, broader implications, execution record, and waiver rationale.

How Advance My Profile approached the case

Advance My Profile treated the matter as evidence based professional development rather than a publicity exercise. The work began with a forensic audit of employer records, data rights, technical authorship, and H-1B constraints. We then helped the client define a narrower specialization, reconstruct two contribution files, prepare permission safe technical works, organize a public research asset, document independent use, and align the U.S. professional plan with the completed record.

The objective was not to manufacture the appearance of expertise. It was to make genuine industrial work attributable, transferable, verifiable, and useful to other professionals. Activities that lacked technical substance, independent verification, lawful execution, or a clear relationship to the endeavor were not pursued.

Frequently asked questions