NM
AI Product Operations / Implementation / Deployment

Nikolay Malyshev

AI Product Operations & Deployment

Turning business workflows into reliable AI operations

I lead the work required to move AI from a business need into daily operation: workflow discovery, product requirements, implementation, evaluation, supervised rollout, operator enablement, and ongoing improvement. Most recently, I built and launched an AI order desk operating with four clients across the United States and Kazakhstan, connecting voice and messaging workflows to ERP data with human review and escalation controls.

Before Allcatch, I spent five years at Amazon Lab126 working within a broader quality organization on performance and functional KPI evaluation for Alexa device programs. I owned delivery across three KPI validation workstreams and contributed to several others, covering validation strategy, recurring execution, failure investigation, and release-readiness reporting. I also restored and stabilized critical validation processes, turning unreliable workflows into repeatable operational programs. Over time, I expanded this work from conventional automation into LLM-assisted user-interface validation, generative AI services, and AI-assisted failure analysis.

Available for full-time work. I have handed off Allcatch's day-to-day operations and am focused on bringing this experience to an AI product operations, implementation, or deployment team.

Explore Production Work LinkedIn Contact
Approach

From business workflow to production use

I begin with the operating process: what users need to accomplish, which data is authoritative, where human judgment is required, and how exceptions should move through the organization.

I turn that model into requirements, acceptance criteria, implementation decisions, evaluation plans, operator guidance, and production controls. My quality background adds disciplined failure analysis, release judgment, and continuous improvement.

Operating Model

How I move AI into operation

  1. 01Discover

    Map the business workflow, user needs, source systems, decision points, and operational risk.

  2. 02Define

    Translate the workflow into product requirements, acceptance criteria, approval rules, and escalation paths.

  3. 03Implement

    Build or coordinate the workflow, integrations, controls, and operating procedures required for use.

  4. 04Launch

    Evaluate known failure cases, prepare operators, and introduce the system through supervised rollout.

  5. 05Operate

    Monitor behavior, investigate issues, support users, and improve the product from production evidence.

Selected Work

AI implementation, modernization, and delivery

Allcatch LLC to Present

Bringing an AI order desk into production

I led Allcatch from customer workflow discovery through production operation for four industrial distribution clients across the United States and Kazakhstan. The system combines voice and messaging intake with ERP data, human approval, and exception management. I owned product direction, workflow design, implementation, evaluation, rollout, operator onboarding, and the operational handoff.

AI product operations / AI deployment / Voice workflows / ERP integration / Evaluation and controls

InputCall or messageMissed, overflow, after hours, or routine request
UnderstandIn-house voice engineCaptures intent and order context
VerifyERP source dataInventory, customer, pricing, and order context
ControlDraft or escalateHuman approval when evidence is incomplete
OperateQueue and monitorReview, recovery, and follow-through
Read the case studyClose the case study

Product and workflow ownership

I translated customer operations into product requirements, source-of-truth rules, approval boundaries, exception paths, and acceptance criteria. I owned the operating model from initial discovery through production use.

ERP integration

The workflow connects inventory, customer records, pricing context, cylinder balances, and order operations. Answers and drafts require verified source data before the system can proceed.

Voice and messaging intake

An in-house voice engine supports first-line calls, missed calls, overflow, after-hours requests, and routine order inquiries. Email and messaging intake follow the same verification, approval, and handoff model.

Deployment and enablement

I defined evaluation scenarios, supervised rollout controls, monitoring, recovery procedures, operator guides, and live walkthroughs. These practices supported adoption while keeping customer-facing decisions reviewable.

Operational handoff

I documented responsibilities, escalation paths, and recovery procedures, then handed off day-to-day operations. I am fully available for a full-time role.

Example acceptance logic

Representative controls used to decide when the system may proceed.

ConditionSystem actionControl
Routine request with verified ERP dataPrepare a grounded draftHuman approval during supervised use
Price or policy cannot be verifiedDo not make the claimEscalate with context
Speech or order field is uncertainFlag the uncertain fieldOperator corrects or confirms
Integration or workflow failureKeep the request visibleHuman handling and recovery procedure

Vendum

Regional implementation for Kazakhstan and Russian-speaking wholesale operations

The same operating model is adapted to 1C data, local workflows, terminology, currency, voice and messaging channels, manager approval, and exception handling.

Core workflowVerify, draft, approve, escalate
Business system1C inventory, pricing, customers, orders
Local operationCurrency, terminology, channels, approval
Amazon Lab126 to

Advancing device quality from automation to LLM-assisted validation

During five years at Amazon Lab126, I worked within a broader quality organization on performance and functional KPI evaluation for Alexa device programs. My work covered validation planning, recurring execution, failure investigation, and release-readiness reporting. I owned delivery across three KPI validation workstreams and contributed to several others. Alongside that work, I advanced selected validation workflows from conventional automation toward LLM-driven user-interface interpretation, generative AI services, and AI-assisted failure analysis.

Performance and functional KPIs / Operational ownership / Automation modernization / LLM-assisted validation / Release readiness

EvaluatePerformance and functional KPIsPlan, execute, analyze, and report recurring validation
StabilizeReliable automationRestore execution and maintain repeatable validation
ModernizeLLM-assisted validationInterpret changing UI states beyond fixed scripted steps
InvestigateAI-assisted analysisUse logs and GenAI services to structure failure evidence
InformRelease readinessCombine KPI results, defects, telemetry, and engineering review
Read the case studyClose the case study

Core focus

My core focus was performance and functional KPI evaluation. I defined validation plans and execution cadence, analyzed results, coordinated partner requests, investigated failures, verified fixes, and provided evidence that informed release-readiness decisions.

Automation foundation

I maintained and extended Python-based automation, stabilized first-use workflows, restored a non-executable validation framework and its supporting environment, and established repeatable KPI execution across new and released device programs.

KPI ownership and breadth

I owned delivery across three KPI validation workstreams and contributed to several others. This work was part of a broader five-year scope covering validation strategy, automation, performance and functional analysis, failure investigation, and release readiness.

Transition to LLM-assisted validation

The Tools team proposed the LLM-based approach and provided a working framework with a limited initial screen set. I owned the practical build-out for selected first-use KPI validation, moving execution from dependence on fixed Appium interactions toward interpretation of visible user-interface states. I onboarded the required screens, stabilized changing flows, improved unexpected-screen handling and logging, submitted code reviews, and initiated a headless execution path intended to reduce the Appium dependency.

AI-assisted investigation

I integrated GenAI API services into automation workflows and built agents that analyzed failure logs before human review. These capabilities reduced repetitive triage and made evidence collection more consistent while keeping engineering review at the center of root cause and release decisions.

Launch evaluation

I supported validation and launch of new Alexa AI capabilities, including Alexa Plus. I combined first-use success, response time, memory, stability, logs, telemetry, and defect evidence to inform release-readiness decisions.

Vimeo to
Zero Motorcycles Contract to

Product quality leadership across media and connected systems

At Vimeo, I led a four-person outsourced QA team, established mobile release workflows, and worked directly with customers and partners on camera and video integrations. In a subsequent contract at Zero Motorcycles, I served as the sole quality owner for a connected iOS and Android product launch spanning Bluetooth, firmware uploads, vehicle telemetry, and app store release.

Team leadership / Release operations / Customer and partner coordination / Connected product delivery

VimeoTeam and workflow leadershipFour-person QA team, mobile releases, camera and video workflows, customer and partner troubleshooting
Zero, contractConnected-product launch ownershipSole QA for iOS and Android release, Bluetooth, firmware uploads, telemetry, and defect coordination
Read the case studyClose the case study

Vimeo

QA Engineer / Apr 2016 to Sep 2018

Team and release leadership

I led a four-person outsourced QA team, established test strategy and release workflows, and coordinated validation across mobile product releases.

Customer and partner integrations

I tested video capture, encoding, playback, and camera integrations. I worked directly with customers, internal teams, and external partners to isolate integration issues and translate findings into product and release actions.

Zero Motorcycles

Software QA Engineer, Contract / Sep 2018 to Apr 2019

Connected product ownership

I served as the sole quality owner for the iOS and Android application launch, validating Bluetooth connectivity, firmware uploads, and real-time telemetry between the application and motorcycle systems.

End-to-end release

I implemented Android UI automation with Espresso, performed iOS compatibility testing, supported App Store and Google Play submission, and managed defects in Jira through resolution.

Experience

Career history

  1. Allcatch LLC

    AI Product Operations & Workflow Builder

    to Present

  2. Amazon Lab126

    Software QA Engineer

    to

  3. Quicken Inc.

    Software Quality Assurance Engineer / Contract

    to

  4. Zero Motorcycles Inc.

    Software QA Engineer / Contract

    to

  5. Vimeo LLC

    QA Engineer

    to

  6. Meshtrip.com

    Software QA Engineer

    to

Capabilities

Capabilities for AI product operations and deployment

AI Product Operations

  • Workflow discovery
  • Product requirements
  • Acceptance criteria
  • Supervised rollout
  • Operator enablement
  • Production monitoring
  • Issue resolution

AI Implementation

  • Voice workflows
  • Messaging intake
  • ERP integration
  • 1C workflows
  • Operational data mapping
  • Human handoff
  • Deployment coordination

Evaluation & Controls

  • Source verification
  • Known failure cases
  • Model comparison
  • Human approval controls
  • Exception escalation
  • Output review

Automation & Reliability

  • Python / PyTest
  • LLM-assisted validation
  • GenAI API integration
  • AI-assisted log analysis
  • Appium / Playwright
  • Selenium / Espresso

Systems & Tools

  • Linux / Ubuntu
  • Device and test labs
  • Wi-Fi / Bluetooth / USB / ADB
  • Logs and telemetry
  • Root-cause isolation
  • Release readiness
  • Jira / Confluence / Jenkins / Git
Contact

Discuss AI product operations and deployment

Allcatch's day-to-day operations have been handed off, and I am fully available for a full-time role. I am interested in AI product operations, implementation, deployment, and enablement opportunities, with selected contract work also welcome.

Connect on LinkedIn Emailn.malyshevqa@gmail.com Production workallcatch.ai
LocationRedwood City, California