Anton Kobchenko

AI Systems & Product Engineer

I turn complex workflows into working products: design the architecture, build interfaces and backend services, integrate AI, and take systems into operation.

AI / LLM systemsProduct architectureFull-stack developmentAgent workflows
1 · AI2 · Web / Full-stack3 · Architecture & systems4 · Science & researchLLMMLBackendDatabasesFrontendInfrastructure & DevOpsProduct designAI / LLM systems: LLM features built around queues, data contracts, workflows and operation.AI / LLMsystemsAgent workflows: Multi-step processes, scenario graphs and tools with explicit data flow.Agent workflowsPrompt engineering: System instructions, reusable prompts and versioned configurations.PromptengineeringStructured output: Response schemas, parsing and typed contracts between models and applications.StructuredoutputLocal LLMs: Private inference with open-weight models, llama.cpp and a dedicated gateway.Local LLMsApplied ML: Applied neural networks, clustering and recommendation logic.Applied MLNeural networks: From early neural network projects to modern LLM applications.Neural networksPyTorch: Training and comparing PyTorch models: a typing-rhythm VAE and flow models for motion trajectories.PyTorchGenerative models: Generating new temporal sequences with VAE and normalizing flows; comparing distributions rather than training loss alone.GenerativemodelsClustering: Grouping episodes by motion characteristics and context to inspect the composition of a research corpus.ClusteringML data preparation: Cleaning public logs, retaining provenance and distribution tails, and splitting data by participant and session.ML datapreparationModel evaluation: Comparing ML model quality through controlled experiments, reproducibility and careful interpretation of results.ModelevaluationRecommendation systems: E-commerce recommendation logic and product research.RecommendationsystemsBackend engineering: APIs, background jobs, integrations and service data.BackendengineeringPython: AI services, FastAPI, research prototypes and infrastructure tools.PythonFastAPI: Typed Python APIs for product and research applications.FastAPIPHP / Laravel: Full-stack and backend development across marketplaces, CRM and platforms.PHP / LaravelQueues & workers: Background execution, retries, load distribution and task hierarchies.Queues &workersKafka: Messaging integration for a medical microservices platform.KafkaPostgreSQL: Relational data models for backend services.PostgreSQLMySQL: Schemas, queries, CRM, marketplaces and data migration.MySQLSQLite: Embedded relational storage for local applications.SQLiteMongoDB: Document storage and application data modeling.MongoDBJS / TS: From jQuery and AngularJS to typed Vue and Nuxt applications.JS / TSVue / Nuxt: Product interfaces, dashboards and server-rendered applications.Vue / NuxtSSR / SEO: Server rendering, search-friendly structure and long-lived product content.SSR / SEOInteractive maps: Visual interfaces for zones, routes and geographic constraints.InteractivemapsLinux / SSH: Server maintenance, health checks and installation workflows.Linux / SSHNginx / load balancing: Gateway routing and distribution of inference requests.Nginx / loadbalancingOrchestration: Repeatable installation, deployment and maintenance workflows across servers.OrchestrationDocker: Isolated recording bots, agents and repeatable service deployment.DockerCI/CD & Git automation: GitHub/GitLab hooks and controlled application updates.CI/CD & GitautomationObservability: Dashboards, telemetry, health checks and task monitoring.ObservabilitySecurity operations: Collection of system events and operational logs for investigation.SecurityoperationsProduct UI / UX: Complex forms, graph editors, dashboards and focused user journeys.Product UI / UXProduct development: Discovery, UX, implementation, feedback and long-term product evolution.ProductdevelopmentProduct architecture: Connecting domain models, user journeys and runtime constraints.ProductarchitectureDomain modeling: Entities, rules and states for complex product domains.Domain modelingGraph structures: Graph models for conversations, AI scenarios and task execution.GraphstructuresMicroservices: Legacy decomposition, service communication and explicit data contracts.MicroservicesEvent-driven architecture: Event-based integrations, message brokers and background processing.Event-drivenarchitectureGeospatial services: Backend services for geographic classification, spatial rules and geodata: from flight approvals to emergency call maps and ambulance movement.GeospatialservicesApplied research: Learning new domains and translating knowledge into working models.AppliedresearchProtein structures: Research prototypes using structural protein data.ProteinstructuresAromatic molecules: Research data, chemical profiles and sources for an aroma product.AromaticmoleculesGeographic regulations: Translating regional rules and approval processes into product workflows.GeographicregulationsOpen full-size graph ↗

Products · systems · infrastructure

Selected work

Independent product · 2025 — 2026

OverSSH — server orchestration

ProjectsHosts Scripts → flows
SSHRun Operate
LogsEvents Package changes
AI analysisRepo context Advice
Projects combine hosts, scripts and flows. SSH orchestration, log monitoring and agent analysis with security recommendations. Orchestration · Monitoring · Proactive security

Hosts, scripts and flows grouped into projects. SSH orchestration, log monitoring, agent analysis of changes and proactive security.

RoleCreator and developer
StackLinux · SSH · Shell scripts · Docker
ResultDeployment, updates and health checks follow repeatable workflows.
Commercial project · 2023 — 2024

Voice AI platform

ContactsPhone database Outbound calls
AI dialogueScript GPT + voice
QualifyQualification By stage
SalesSales call Deal
Contacts enter outbound calls, scripted GPT dialogue and qualification. Qualified leads go to a sales call; others exit. Funnel analytics and A/B testing. Stage conversions · A/B scripts

Calls from a contact database, scripted dialogue with GPT and lead qualification. Qualified leads passed to sales, stage-by-stage drop-off analytics and A/B testing of scripts.

RoleLead full-stack / AI developer
StackPython · GPT API · Cloud STT/TTS · Laravel
ResultBuilt a platform with a voice agent, scenario editor, CRM features and analytics.
Research · October 2026

Human Input: motion & typing

DataPreparation Scenarios
EvaluationQuality Variation
GenerationMovement Typing rhythm
Input planAPI Interactive demo
A generator of mouse movement and typing rhythm with natural variation. Turns a target and text into a ready-to-use sequence of actions, with an API and an interactive demo. Mouse movement · Typing rhythm · EN / RU / Unicode

A generator of mouse movement and typing rhythm with natural variation. Turns a target and text into a ready-to-use sequence of actions, with an API and an interactive demo.

RoleAuthor and developer
StackPython · PyTorch · NumPy / SciPy · PyArrow
ResultBuilt a generator of mouse movement, typing rhythm and ready-to-use input plans with an API.
Commercial project · 2024

Medical platform microservices

PatientsRecords Client data
GeospatialCalls and movement WebSocket
SettingsShared configuration For all services
Three extracted services: patient records, geospatial updates over WebSocket/Centrifugo and shared settings. Kafka messaging and PostgreSQL data. PHP · Kafka · PostgreSQL · Centrifugo

Patient records, shared settings and geospatial services for calls and ambulance movement. Realtime map updates via WebSocket/Centrifugo, with Kafka messaging between services.

RoleBackend engineer
StackPHP · Kafka · PostgreSQL · Microservices
ResultImplemented patient, configuration and geodata services, plus a shared Kafka integration module.

From idea to working solution

More projects

Reading the scales

Two defining aspects of each project. Three segments indicate their extent based on the case study; these are the author's qualitative assessments.

  • Complexity: Local tasks → Connected subsystems → Complex system and constraints.
  • Creativity: Adapting existing solutions → Custom approach → Original model or mechanics.
  • Research: Learning the tools → Domain investigation → Experimental model and hypothesis testing.
  • Integrations: One external system → Several systems or platforms → Multi-party product workflow.
  • Ownership: Individual features → End-to-end development and delivery → The whole product and business.

Taking on more complex problems

Professional journey

  1. 2008 — 2012

    Mathematics, physics & programming

    NSU specialized school and Faculty of Physics: strong foundations in mathematics, algorithms and early neural-network projects.

  2. 2012 — 2018

    From games to full-stack

    From C++ and mobile games to end-to-end web products: interfaces, JavaScript, PHP, a marketplace and a forms service.

  3. 2018 — 2021

    Building a product and a business

    Legal Drone: market research, geodata, approval workflows and a paying customer.

  4. 2021 — 2022

    Working across a large team

    Product features across design, frontend, backend, QA and release.

  5. 2023 — 2024

    Voice agents & scenario graphs

    Full-cycle AI platform development with dialogue editing, versions and analytics.

  6. 2025 — 2026

    AI systems & infrastructure

    Meeting recording, a shared LLM task engine, multi-step workflows and local-model operations.

Let’s talk about your project

AI features, a new product or an existing system ready for its next step — let’s talk.

Products, systems and ideas brought to life in working software.