Selected work & AI systems

I turn complex operations into software people can understand, trust, and use.

I’m Yuri Samoilovski, founder of TaskGeniusAI. I help businesses bring disconnected systems, workflows and data together so their operations become clearer, more efficient and intelligently automated.

My perspective comes from both sides: building and operating businesses across multiple sectors, partnering closely with other business owners, and spending roughly 15 years engineering production systems across payments, APIs, infrastructure, automation and applied AI. That combination helps me understand how a business actually operates — and build technology around the operation, rather than forcing the operation around the technology.

See selected workView TaskGeniusAI
Selected Work & AI Systems | TaskGeniusAIData, workflows, people, rules, decisions and existing software feed one connected layer holding shared context, evidence, permissions and operational state. Out of it come clarity, control, attention and usable action.INPUTSCONNECTED LAYERRESULTSDataWorkflowsPeopleRulesDecisionsExisting softwareShared contextEvidencePermissionsOperational stateClarityControlAttentionUsable action
  1. InData · Workflows · People · Rules · Decisions · Existing software
  2. ConnectedShared context · Evidence · Permissions · Operational state
  3. OutClarity · Control · Attention · Usable action

Engineering rigor. Operator perspective.

My background combines two perspectives that do not always exist in the same room. I have spent roughly 15 years building production systems where reliability, accuracy, security and failure handling matter. I have also operated businesses myself, where software is judged by whether it makes the work clearer, more dependable and easier to control.

That combination shapes how I approach AI: understand how the work moves, establish what can be trusted, and then decide where intelligence belongs.

Experience building and operating businesses across different commercial and customer-facing models — working directly with acquisition, fulfillment, payments, customer experience, staffing, financial operations and growth.

Built and operated a business that exceeded $1 million in first-year revenue.

Selected Work & AI Systems | TaskGeniusAIFive paired relationships meet in a shared decision layer: reliability with customers, accuracy with money, architecture with workflow, integration with decisions, and failure handling with growth pressure. Influence runs in both directions, and together they resolve into systems that work in reality.ENGINEERINGDECISIONSOPERATINGReliabilityCustomersAccuracyMoneyArchitectureWorkflowIntegrationDecisionsFailure handlingGrowth pressureSystems that work in reality
  • ReliabilityCustomers
  • AccuracyMoney
  • ArchitectureWorkflow
  • IntegrationDecisions
  • Failure handlingGrowth pressure
  • TogetherSystems that work in reality

Four disciplines, one standard.

These are not four separate careers. Each one keeps correcting the others, and they are all judged against the same thing.

Production engineering

Build for accuracy, resilience, integration and long-term operation.

  • APIs and integrations
  • Production infrastructure
  • Data processing
  • Failure handling

Product building

Turn complex capabilities into understandable tools and workflows.

  • Workflow design
  • Reporting
  • Reconciliation
  • Information design

Business operations

Judge the system by how it performs during real work and real pressure.

  • Operational feedback
  • Measured outcomes
  • Cost and margin reality
  • Day-to-day usability

Applied AI

Place intelligence inside a defined workflow with evidence, review and accountability.

  • Structured extraction
  • Human review
  • Permissions
  • Evidence and validation

Judged byUseful, trustworthy systems

Enterprise AI-assisted engineering workflow

Large specifications, made reviewable.

Some of my enterprise work has focused on turning large technical specifications into structured, reviewable engineering inputs. The goal is not simply to generate text. It is to reduce repetitive interpretation while keeping validation and engineering judgment at the point where output becomes a real decision.

Every extracted requirement keeps a line back to the clause it came from, validation runs before anyone is asked to judge the result, and review can send work back rather than only wave it through.

The same enterprise work has also included automating how work moves through large organizations — carrying support and engineering requests through intake, classification, routing, escalation, developer assignment, implementation, QA and code review instead of hand-coordinating each step. Across pipelines that span several teams, that removes a great deal of repetitive coordination.

Details generalized to protect confidential work. Client identity, proprietary architecture, internal metrics, source code and confidential materials are intentionally excluded.

Selected Work & AI Systems | TaskGeniusAIThree clauses are extracted from a dense specification into structured requirement records, each keeping a line back to its source clause. They become one structured interpretation, pass source-match, completeness and conflict checks, go to a human engineering review that can approve or return the work, and end as a usable engineering output.SPECIFICATION§ 2.1§ 3.2§ 4.3§ 5.4R-01Requirementtraced to § 2.1R-02Requirementtraced to § 3.2R-03Requirementtraced to § 4.3Structured interpretation3 requirements · source anchors keptValidationSource matchCompletenessConflictsHuman engineering reviewRequired before anything shipsApproveReturnUsable engineering output
  1. Technical specification
  2. Requirements extracted, each traced to its clause
  3. Structured interpretation
  4. Validation — source match, completeness, conflicts
  5. Human engineering review — approve or return
  6. Usable engineering output

Building TaskGeniusAI.

TaskGeniusAI grew from a pattern I kept seeing across operating businesses: the company already had software, but its activity, money, work and next decisions lived in different places. The product connects those records into one operating picture and keeps what it knows — and what it does not yet know — visible.

  1. ConnectBring relevant systems, activity, financial records and workflow events into one operating context.Operational sourcesFinancial relationshipsActivity and workflow events
  2. UnderstandReconcile identity, lifecycle, relationships, source health and business state.Identity and lifecycleReconciliationSource healthData trust
  3. Act with contextSurface attention, recommendations, decisions and outcomes inside defined controls.AttentionEvidenceRecommendationsPermissionsOutcomes

The packaged Platform is built first around service operations. The same connected-systems approach can be applied selectively to other complex operational workflows through Custom AI Systems.

What that looks like in the product.

Synthetic demo workspace — no customer data

TaskGeniusAI Performance Overview for the Juniper Coast demo workspace on a last-thirty-days window, headed Demand is growing and marked Full visibility with acquisition and job data both current: 161 real inbound leads up 10 percent, 103 booked jobs up 27 percent and 83 jobs completed up 6 percent, a whole-business conversion funnel, paid-acquisition efficiency at 84 percent attribution coverage, and a demand-and-booking trend chart compared against the previous period.
One connected operating picture. Spend, demand, booked work, completed work and revenue read from the same model over one declared window — with the trend beside the numbers and the reason the period moved stated in plain language.
An evidence panel in TaskGeniusAI headed Work the recovery queue before changing the intake process, listing the supporting facts behind it, labelling booking evidence partial and recovery evidence confirmed, noting that first-response time is unavailable, and offering a Review leads action beside a link to the supporting evidence.

Attention you can inspect

An item states what was noticed, the records behind it, and what it cannot yet support — booking evidence partial, recovery evidence confirmed, first-response time unavailable. The proposed next move sits with the evidence, not in place of it.

TaskGeniusAI Financials for the Juniper Coast demo workspace: a financial data trust row showing full coverage and current freshness on completed revenue, invoice balances and payment dates, 97 percent on ad spend and none on labour or parts costs; a revenue activity chart; receivable ageing totalling $60,452 across 40 open invoices; and a profit readiness table marking gross margin and operating profit unavailable until five named cost inputs are connected.

Numbers that stay reconciled

Every figure carries its coverage and its freshness, and receivables stay separate from revenue because their underlying status differs. Operating profit is not estimated to fill the gap: it reads unavailable, and the page names the five cost inputs required to compute it.

TaskGeniusAI data coverage for the Juniper Coast demo workspace: payments marked partial because 206 of 246 invoices carry a payment date, and labour costs marked missing because no integration supplies them yet.

Honest about what it knows

Current, stale, partial and missing are four different states, and the system says which one it is looking at rather than resolving them all into a confident number.

A technically impressive system still fails if the people operating it cannot understand it, trust it, or use it when the work gets busy.

How I approach the work.

  1. 01Understand before automatingLearn how the work actually moves before deciding where AI belongs.
  2. 02Build on what is valuableExisting systems and prior investments should be understood before replacement is considered.
  3. 03Keep evidence attachedImportant numbers and recommendations should remain connected to their source.
  4. 04Make uncertainty visibleMissing, stale and conflicting data should not become confident answers.
  5. 05Keep control intentionalAI and automation should operate within clear permissions, review requirements and business boundaries.

If this overlaps with what you’re building, let’s compare notes.

No prepared solution — just a useful conversation about the systems, workflows and decisions behind the work.