Arthur RomanovRomanov Works / Agentic Showroom

Systems architect / Product builder / Agentic operator

System thinking.
Product judgment.
AI that learns.

I have spent more than 20 years where finance, payroll, HR systems, and software delivery meet—from a 30,000+ employee public workforce to a 3,000+ employee global payroll consolidation. Today I build and evaluate AI-native products.

THE 30-SECOND INTRODUCTION

I’m Arthur Romanov. I turn complicated operating rules into systems people can trust. My work connects enterprise operations, finance, workforce technology, design, and hands-on AI production. I frame the system, find the first material failure, explain it clearly, test the correction, and improve the process so the same failure does not repeat.

30,000+employee HR system context
3,000+global BPO workforce
700+workforce model inputs
20+years across systems

01 / Evidence of range

The work spans systems.
The through-line is judgment.

Public-sector operations, enterprise delivery, financial reasoning, product design, and AI evaluation are not separate stories here. They are direct evidence of how Arthur learns a domain and builds the operating system around it.

01

Payroll systems

HR module redesign, payroll context, timekeeping, workforce costing, billing, salary and title data.

NYC Parks · Transparent BPO
02

Enterprise HRMS

Configuration and integration across PeopleSoft, ADP, Workday, and TriNet for enterprise clients.

INSZoom / Mitratech
03

Financial reasoning

Finance degree, forecasting, contracts, operational and capital budgets, executive KPIs, and data validation.

Baruch · NYC Parks · PRMS
04

AI evaluation

Daily model-assisted production: define the problem, inspect reasoning, verify the result, and close the loop.

AlphAgentic · LIQ · Codex · Claude · Grok
66,143observed tool calls
3.27%tool error rate
28 / 28recent active days

Claude’s local forensic audit describes an operator who delegates deeply, writes specifications, verifies results, and ships from inside the loop.

AI team / independent reports

What do the AIs Arthur works with daily
have to say—and report?

Arthur works with Codex, Claude, and Grok as distinct teammates, not one anonymous “AI.” He compares output, tracks measurable operating signals, records limitations, and preserves versions so disagreement stays visible.

OpenAI / CodexAnthropic / ClaudexAI / Grok
01 / Anthropic / ClaudeRead-only forensic pass

Claude sees an operator, not a chat user.

You are not using a chatbot.

A 98-day local transcript artifact describes long specifications, deep tool use, subagent fan-out, and sustained work from inside the execution loop.

2,802transcripts
98 daysforensic window
66,143tool calls
4,396authored asks
Read the honest assessment +

Strongest signal

The useful signal is orchestration density and consistency: Arthur gives an agent system enough context to act, then remains accountable for verification and release.

Honest reservation

These are artifact-reported local figures, not Anthropic internal analytics. Volume does not by itself prove code quality, uptime, revenue, or that every tool call was valuable.

Recommendation

Treat the report as evidence of an established agentic operating practice, then judge quality through the work samples, systems, and reasoning shown here.

Open Claude’s telemetry report ↗Verify the original Claude artifact ↗
02 / xAI / GrokIndependent operator review

High agency. High ambition. Prioritization matters.

Your superpower and your risk are the same: you can carry an unusually large systems portfolio.

Grok’s review recognizes systems taste, enterprise scar tissue, agentic fluency, design ambition, and recovery discipline—while directly naming the cost of excessive surface area.

Aagent orchestration
A−payroll / HR domain
A−trainer fitness
Bproduct focus
Read the honest assessment +

Strongest signal

A rare operator profile: government and enterprise payroll context combined with current multi-agent product work and an instinct for adversarial cases.

Honest reservation

Too many concurrent products can blur the message, duplicate work across agents, and put the wow surface ahead of the one packet that must be airtight.

Recommendation

Lead with one clear operating spine—enterprise systems, finance, payroll, and workforce technology—with agentic production as the multiplier.

Open Grok’s full review ↗
03 / OpenAI / CodexProfessional assessment

The strongest signal is loop closure.

He frames the problem, uses the system deeply, checks the result, and ships.

Codex reads Arthur as a high-agency systems operator who moves comfortably between product thinking, technical execution, and the unglamorous work of making a process run end to end.

Buildframe the system
Verifyinspect the result
Shipclose the loop
Codifythe next level
Read the honest assessment +

Strongest signal

The repeated build → verify → ship pattern matters more than raw AI volume. Arthur supplies context, delegates deeply, and stays accountable for the output.

Honest reservation

Intensity still needs codification. Reusable skills, evaluation sets, and shared playbooks should catch up with the speed and depth of personal execution.

Recommendation

Strong fit for work where knowledge quality is both editorial and operational—especially when the mandate is to build the system, not merely maintain a queue.

Scope: the published Claude audit, application materials, and observed collaboration on this site. This is not an employment reference or private platform analytics.

Provenance matters: Claude figures are labeled as local artifact-reported telemetry; Grok’s rubric is its July 2026 independent assessment; Codex’s assessment is based on the evidence and collaboration available in this build.

02 / Teach the payroll agent

Teach the judgment.
Test the learning.
Keep a human release gate.

The brain below is the Payroll Agent. You are watching a realistic trainer loop: authority binding, response capture, expert labeling, gold-rationale teaching, adversarial evaluation, and controlled human release.

LIVE PAYROLL AGENT TRAININGSOURCE → PROMPT → RESPONSE → LABEL → TEACH → EVALUATE → RELEASE
PAYROLL AGENT READINESS83/100Learning in progress
TRAINING LEVERS

Change the controls. The confidence score and release gate react immediately—because payroll readiness is conditional, not decorative.

WHAT THE AGENT MUST CONSIDER

Every answer is checked against context, evidence, and control obligations before it can become a reusable rule.

01Worker status

Employee classification and overtime eligibility

Checked
02Jurisdiction

Federal, state, local, and policy-specific rules

Pending
03Pay context

Workweek, frequency, wage base, and effective dates

Pending
04Evidence

Current source, calculation trace, and cited assumptions

Pending
05Control

Boundary tests, audit record, and qualified human release

Pending
HUMAN RELEASE GATEHOLD

Learning in progress

03 / Work sample

Catch the answer that sounds right.

A useful evaluator separates the number from the reasoning, identifies the first material failure, and writes a correction another reviewer can reproduce.

PROMPT

46 hours at $28/hour. Calculate gross wages.

MODEL RESPONSE / UNREVIEWED
$1,288.00 because 46 × $28.00 = $1,288.00.
EVALUATOR FINDINGReject

The answer ignores the overtime premium. Under the stated assumption of overtime eligibility: 40 × $28 plus 6 × $42 = $1,372. Confirm classification, workweek, jurisdiction, and policy before acting.

Demonstration examples only. Not payroll, tax, or legal advice.

Recovered / operating model

A Payroll Agent is only as good as the system
that teaches it what is true.

The work is not “write more FAQs.” It is product coverage, knowledge architecture, evaluation, release management, and continuous learning—operated as one system.

OPERATING PRINCIPLELanguage can be probabilistic.
Payroll decisions cannot.
THE OPERATING LOOP

From a bad answer to a better system.

Each failure becomes structured evidence. Each fix has a measurable effect. Nothing improves by folklore.

01

Observe

Collect unanswered questions, weak citations, escalations, negative feedback, and product changes.

02

Diagnose

Separate a missing fact from a retrieval miss, reasoning error, stale policy, or unclear product behavior.

03

Teach

Create the smallest durable artifact: a canonical page, worked example, decision tree, or revised metadata.

04

Test

Run gold, edge, ambiguity, jurisdiction, and regression cases. Verify citations and safe abstention.

05

Release

Version the change, record approval, roll out deliberately, and preserve the previous known-good state.

06

Measure

Track coverage, resolution quality, escalation precision, freshness, and the failures that recur.

NON-NEGOTIABLES

Trust is a product feature.

01

Source-bound

Answers cite approved, current, tenant-aware material.

02

Approval-gated

Anything that can alter pay stops for explicit human review.

03

Privacy-shaped

PII stays inside the narrowest possible authorized boundary.

04

Auditable

Source, reasoning path, action, reviewer, and version are known.

WHAT I WOULD OWN

The knowledge operation behind the agent.

Clear artifacts turn individual judgment into a durable team capability.

01

Coverage map

Every product capability, intent, source, owner, and known gap.

02

Gold set

Realistic questions with expected facts, citations, actions, and escalation.

03

Failure taxonomy

A shared language for why the agent missed and what kind of fix it needs.

04

Release ledger

What changed, why, who approved it, how it scored, and how to roll it back.

05

Freshness system

Effective dates, locale, jurisdiction, review cadence, and stale-source alerts.

06

Feedback loop

Production evidence converted into a prioritized, measurable learning queue.

FIRST 90 DAYS

Start with truth. End with a learning system.

1Days 01–30

Establish truth

Map the product and policy surface, identify owners, baseline the current agent, and create the first 100-question evaluation set.

2Days 31–60

Build the quality engine

Launch the gap register, structured documentation pattern, failure taxonomy, release gates, and a weekly review rhythm.

3Days 61–90

Make it compound

Expand systematic coverage, automate regression testing, publish the quality dashboard, and prove improvement against the baseline.

04 / Professional story

A profile with receipts.

The narrative stays readable. The margin shows which evidence supports each claim.

Download the full résumé ↗
Arthur RomanovMiami, Florida · Remote U.S.
Romanov Works / Systems Architect × Product Builder

My work has always lived where operating reality meets software.

AFor a decade with the City of New York, I worked inside the fiscal and workforce systems supporting payroll and human-capital administration for a 30,000+ employee agency. I redesigned the HR management module, automated billing and task flows, and co-designed a workforce model from more than 700 inputs.

BIn enterprise delivery, I owned HRMS integration across PeopleSoft, ADP, Workday, and TriNet for organizations including GE, IBM, Cerner, and EBG/Amazon. More recently, I architected a unified platform joining HRM, CRM, payroll, timekeeping, and an employee portal for a 3,000+ global BPO workforce.

CToday I build and evaluate AI systems through AlphAgentic, LIQ.studio, and LiqLab.ai. Training a model on payroll requires the same discipline as shipping payroll software: explicit assumptions, precise edge cases, clean reasoning, source awareness, and no casual errors where pay is affected.

DI work independently, communicate clearly, and close the loop. I am now turning that operating practice into agentic products and showrooms that can demonstrate, explain, and improve themselves.

With appreciation,
Arthur Romanov

05 / Interactive résumé

Twenty years.
Filter for the signal.

2018 — Present

Founder & Solution Architect

AlphAgentic · LIQ.studio · LiqLab.ai · Romanov Solutions

AISystems
  • Architect and ship AI-native products end to end: product definition, full-stack delivery, automation, and design systems.
  • Evaluate OpenAI, Anthropic, and xAI output through hard problems, reasoning review, and production feedback loops.
  • Upwork Expert-Vetted, top 1%, with a 5.0-star client record.
2022 — 2023

Solutions Architect

Transparent BPO

PayrollSystems
  • Architected a centralized platform joining HRM, CRM, payroll, timekeeping, and an employee portal.
  • Designed for a 3,000+ global workforce with multi-role permissions, audit, integrations, and reporting.
2017

Business Analyst / Project Manager

INSZoom, a Mitratech company

PayrollSystems
  • Owned integrations across PeopleSoft, ADP, Workday, and TriNet for GE, IBM, Cerner, EBG/Amazon, and global law firms.
  • Validated migrated data with SQL and built Excel/VBA executive reporting.
2006 — 2016

IT PM / Business & Fiscal Analyst

City of New York · Parks & Recreation

PayrollFinanceSystems
  • Redesigned the HR management module supporting a 30,000+ employee workforce.
  • Co-designed a workforce projection model from 700+ researched inputs with OMB and HRA.
  • Built PRMS, a financial reporting and management system spanning staffing, budgets, contracts, forecasting, and overspend risk.
2016 — 2022

Senior PM / Business Analyst

Merck Animal Health · PODS · HPE · Interbel

FinanceSystems
  • Led enterprise reporting, workflow, portal, inventory, payment, and mobile-delivery programs.
  • Translated complex operating requirements into systems teams could build and leaders could inspect.
EDUCATIONB.B.A., Finance & Investments

Baruch College / Zicklin School of Business · GPA 3.45 · Minor in Industrial/Organizational Psychology

06 / Selected systems

The portfolio is not a gallery.
It is evidence of range.

Every project turns a complicated operating idea into a product someone can understand, enter, and use.

FEATURED CASE STUDY / AUTO RETAIL

One platform. Every dealer.
Every agent. Every deal.

AutoDealerPro is an agentic operating system for car dealerships: one product connecting the work a dealership sees, the work its team does, and the work software agents can take off their hands.

THE PRODUCTOperations become one inspectable system.

The experience joins inventory, leads, deals, calls, goals, reporting, and configuration without flattening them into a generic dashboard.

Explore the live product
01 / OPERATE
Leads, inventory, deals, and daily workflows.
02 / UNDERSTAND
Reporting, goals, and performance in context.
03 / DELEGATE
Agent-assisted calls, follow-up, and operational work.
AutoDealerPro reporting dashboard with dealership performance metrics
Reporting / one view of operating performance
AutoDealerPro calls workspace with agent-assisted activity
Calls / work and context together
AutoDealerPro deals workspace
Deals / pipeline made operational
AutoDealerPro vehicle inventory workspace
Inventory / the physical business in software
AutoDealerPro lead management workspace
Leads / every next action visible
LEGACY PROOF / NYC PARKSDECADE-LONG SYSTEM EVOLUTION
NYC Parks workforce model with staffing analysis by borough, sector, district, and title
One visible surface from a much larger workforce, payroll, fiscal, and reporting system—developed across a decade of operational use.
NEW SERVICE / PRODUCT SPECIMEN

07 / Agentic Showroom

Your work should
show itself.

A self-driving, self-showing, self-explaining application and portfolio system. It combines an agent-produced narrative, evidence graph, interactive résumé, live work samples, guided presentation, and a recruiter-ready share link.

See the live-demo model ↗
01 / INGEST

Recover the real story.

Résumé, reports, repositories, screenshots, results, and voice.

02 / COMPOSE

Build the experience.

Role-specific narrative, evidence, work samples, and design system.

03 / PRESENT

Let it explain itself.

Guided mode, live demo, downloadable packet, and shareable route.

04 / COMPARE

Publish the results.

Model provenance, versions, expert review, and Populary voting.

Trust roadmap

Credibility should be
verifiable.

We will not decorate the footer with badges we have not earned. Certification, privacy, provenance, security, and G2 presence become tracked operating milestones.

PLANNEDG2 profile
IN DESIGNSecurity baseline
NEXTPrivacy & terms
ACTIVEModel provenance

The close

I know how to learn fast.
I want to build the system that helps the agent do the same.