Software Architect · Technical Lead
Business systems · Data at scale · AI infrastructure
I turn complex business operations into working systems.
Start with the real process: people, documents and business constraints.
Define entities, relationships, rules and the boundaries between systems.
Build APIs, workers and integrations into a system that can grow.
Observe what happens, resolve bottlenecks and keep the product working.
I model the business, design the system, write the backend and take it through delivery.
Real problems. Practical solutions. From messy processes to systems that scale.
View all workCommerce operations had to stay consistent across different product models, warehouses and external systems.
Lead Developer · architecture, implementation and delivery
The work connected the product lifecycle from catalog and inventory to orders, fulfillment and tracking, with ownership extending beyond individual services.
Read caseA business operations platform, starting with mortgages
I designed and developed a Python operations platform combining dynamic documents, formulas, workflows, communications and employee compensation around a mortgage business.
Mortgage operations involved linked data, documents, people and actions that a simple task list could not model.
Business modelling and system architecture
Mortgage workflows gave the platform its first concrete business use case. The reusable core was deliberately designed to support other organizations and processes through configuration, documents and rules.
Read caseI built a working AI operations service that connects agents to project data, logs, communication and tasks through controlled tools and reusable project knowledge.
Operational questions span several systems, while broad database access and an ever-growing tool catalog create permission and context problems for agents.
Agent infrastructure and integration architecture
MCP Hub is used in real work: reviewing messages, checking errors, querying project activity and tracing a user journey across systems. The service is currently private, with reusable project knowledge and controlled execution at its core.
Read caseLarge-scale Amazon scraping and trend analysis
I designed and built an Amazon data service that scanned categories across the marketplace, excluding books, and used Elasticsearch bulk processing to identify trends.
The breadth of the catalog required parallel acquisition and a storage strategy that could work with millions of listings without turning every item into a separate read or write.
Data acquisition and processing architecture
The service ran for about five months at a historical scale on the order of millions of listings. It supported category-wide trend analysis before the team changed focus and the project was archived.
Read caseTechnical depth, business understanding, and hands-on execution.
Catalogs, marketplaces, warehouses, orders and fulfilment: the shared model and the boundaries between them.
Designed shared product, inventory and order models for Onlihub and MySellerHub, connecting channels with different warehouse models and external identifiers.
Connect products, SKUs, variants and listings while preserving each channel’s identifiers and rules.
Connect stock, reservations, customer and supplier orders, shipping and tracking across warehouses.
Handle repeated webhooks, rate limits and partial failures with retries, idempotency and reconciliation.
Mortgage operations built around dynamic documents, relationships, permissions, formulas, stage-based automation, email and employee compensation.
Designed BOS around documents with state, formulas and consequences: a stage can trigger a calculation, email, task or financial movement. ERP experience informs the model.
Connect organizations, departments, memberships and permissions to document types, relationships and workflow stages.
Combine nested fields, formulas, calculators and print templates with stage-based tasks, notifications and external mailboxes.
Tie hourly and piece-rate compensation, transactions and document movements to explicit rules and an inspectable history.
Large-scale parsing with concurrency and proxy pools; Elasticsearch bulk reads and writes; separate database, cache, read and worker paths.
Amazon data: parallel collection and Elasticsearch batch reads/writes to reduce storage calls. Catalog: fast Rust reads, Python fallback and Redis caching, with traffic control.
Coordinate asyncio, processes and threads, account/proxy pools, throttling and retries for large catalogs.
Normalize and deduplicate data, then batch Elasticsearch indexing and reads to avoid thousands of small storage calls.
Separate transactional data, search/read paths, Redis cache and background workers; diagnose bottlenecks with query plans and logs.
MCP Hub connects agents to databases, logs, tasks and communication through tool discovery, scoped permissions, approvals and compact context.
Designed MCP Hub around controlled access to databases, logs and work tools: project context, approval points, audit, reusable knowledge and automations that connect several systems.
Retrieve relevant tools and schemas on demand, keeping the visible interface and context focused on the task.
Define permission boundaries, approval points and repeat-operation behaviour before tools change real systems.
Use bounded results, stored outputs, receipts and state inspection to understand what actually happened.
Technical ownership combines business understanding, architecture, hands-on implementation and team coordination.
Took over Onlihub architecture, database, backend and integrations after the lead developer left. Coordinated the team and advised the CEO on feasibility and implementation.
Work with the business to understand the process, its constraints and what a useful result would change.
Translate the model into database, backend and integration decisions; explain the choices while implementing them.
Coordinate tasks and development, handle production problems and keep improving the product after launch.
Manufacturing, CAD/CAM and CNC built the foundation: understand materials, operations and constraints before automating them.
Prepared parts, machine programs and technical documentation for serial production; refined operation sequences and material use, then automated production planning and accounting in 1C.
Model parts and assemblies, choose manufacturing operations and account for materials and tooling.
Prepare machine programs, postprocessors and technical documentation for cutting, turning, milling and punching.
Refine operation sequences, material consumption and bottlenecks as products move into serial production.
Independent product work across game backends, TON, Unity physics, modular vehicles, procedural worlds and 3D assets.
Led TON Tanks product development and built its backend and game logic. Independent Unity work explores modular tanks, tracked physics, damage and procedural city generation.
Connect game rules, progression, rewards and ownership with the backend and product concept.
Explore modular vehicles, damage, tracked physics, multiplayer and procedural environments in Unity.
Create and optimize models, materials, textures and engine audio banks for use in a real-time game.
Before software, there were real machines, materials and production lines. That understanding still shapes how I build.
Explore my backgroundParts, machine programs, materials and serial production. Engineering began with physical constraints and the consequences of each operation.
Working with customers in 1C connected production, stock, payroll, logistics and documents. Software had to reflect what the business actually did.
Commerce platforms brought together catalogs, channels, warehouses, payments and fulfilment, with synchronization across systems that use different models.
Catalog read paths, data pipelines and business operations required different models: transactions, search, cache, dynamic documents and workflow rules.
At Onlihub, I took over technical ownership after the lead developer left: architecture, database, backend, integrations and coordination of a small cross-functional team.
MCP, tool discovery, context management and local AI extend the same approach: give agents a useful view of real systems, defined permissions and inspectable execution.
A role, a product, or a technical challenge. Share a little context.
To use the form, enable JavaScript. You can also write to me directly on Telegram.
Write on Telegram