I architect and ship production-grade AI systems at scale. 4+ years of hands-on ML/AI engineering, system design, and technical leadership across two companies and a team of ~10 engineers.
I build at the intersection of AI research and production reality: LLM observability platforms, multi-agent orchestration (50+ node LangGraph pipelines with conditional routing and human-in-the-loop), dual-database architectures (OLTP + OLAP), and ML pipelines that handle multi-TB workloads in the real world.
Track record: 20+ delivered business projects. 30+ system architectures designed from scratch. Manufacturing, civic tech, e-commerce, industrial sectors.
Co-Founder & CTO/CAIO | Building the debugging layer for the AI era
Curestry is an AI operations platform providing intelligent diagnostics for LLM-powered applications. It combines automated root cause analysis, systematic prompt optimization, and high-throughput trace analytics into a unified observability layer.
Key differentiators:
- 15 specialized analyzers across a 4-tier analysis pipeline (42 metrics total)
- Multi-agent RCA - LangGraph 12-node orchestration for automated failure diagnostics
- Prompt Optimization - A/B testing, scoring, and auto-tuning with multi-vendor LLM support
- Client ecosystem - VSCode extension + Chrome extension (MV3) for in-workflow integration
- Dual-database architecture - PostgreSQL (OLTP) + ClickHouse (OLAP) for traces at scale
Curestry Platform (11 services)
├── Web Dashboard Next.js 16, React 19, TypeScript, tRPC
├── Worker Service BullMQ job processing, background analytics
├── RCA Core FastAPI, LangGraph, LiteLLM (Vertical Slice Architecture)
│ ├── 8 feature slices (findings, analysis, code_scanner, comparator, chat, connectors, optimizer, systems)
│ ├── Multi-agent RCA Root cause analysis with corrective recommendations
│ └── Findings engine 51 types across 10 categories, SHA256 deduplication
├── Prompt Optimization 15 analyzers, 4-tier pipeline (quick→standard→thorough)
├── Client SDK Python SDK + JavaScript SDK (@curestry/client, @curestry/core, @curestry/langchain)
├── Extensions VSCode extension + Chrome extension (Bridge pattern + FSD)
├── Data Layer
│ ├── PostgreSQL 18+ OLTP - Prisma + Alembic migrations (strict ownership)
│ ├── ClickHouse 25.8 OLAP - golang-migrate, ReplacingMergeTree, time-series optimized
│ ├── Redis (x2) Queue (:6379, BullMQ, noeviction) + Cache (:6380, allkeys-lru)
│ └── MinIO S3-compatible blob storage
├── Observability Prometheus + Grafana + Loki
└── Infrastructure Docker (11 services), Caddy reverse proxy, Turborepo + pnpm
Co-Founder & CEO | nddev.it.com
Full-cycle AI/IT outsourcing and outstaffing studio (~10 engineers). We design and deliver production solutions for manufacturing, civic tech, and enterprise clients.
Six divisions: NDDev Dev, NDDev AI, NDDev Design, NDDev Platform, NDDev RnD, NDDev OpenNetwork.
Delivery focus: ML systems, computer vision, full-stack platforms, multi-agent services, mobile applications.
Client projects delivered by NDDev. Details anonymized.
| Domain | Scope | Architecture Highlights | Scale |
|---|---|---|---|
| City-wide digital library ecosystem | Mobile + Admin + Backend | Flutter (Riverpod), React, FastAPI. Two-service backend, Meilisearch, Firebase, biometric auth, barcode integration | 123 endpoints, 25 models, 1436 tests, 29 screens, 38 admin pages |
| B2B industrial equipment platform (UAE) | Full-stack bilingual catalog (EN/AR) | Next.js ISR/SSG, async FastAPI, Feature-Sliced Design, GitHub Actions CI/CD | 10+ domain modules, zero-downtime deploys |
| Industrial equipment sales platform | Corporate site + Admin + API | Layered architecture, React Admin, Redis rate limiting, 2FA, SEO with JSON-LD | 541+ products, 13 service modules |
| Computer vision for road infrastructure | ML pipeline + CV | Real-time object detection, edge deployment | Production safety system |
| ML systems for mining operations | Data pipeline + ML | Multi-TB data processing, optimization models | Industrial-scale analytics |
| CV for collaborative whiteboards | Real-time CV pipeline | Object recognition on canvas, Miro-like platform integration | Real-time processing |
| ML system for agricultural machinery | Embedded ML | Smart harvester control systems | On-device inference |
| Subscription marketplace (CIS) | Frontend platform | Next.js, ISR, i18n, Server Components | Region-wide marketplace |
hackathon-ai-auditor-agent - AI-powered code auditing platform with multi-agent analysis. FastAPI + LangGraph + Next.js + VSCode extension + Chrome extension + Admin panel. Monorepo (pnpm workspaces), Docker Compose, PostgreSQL, Redis. NFNG Hackathon.
local-llm-prompt-optimizer - Offline prompt A/B testing and auto-tuning for local LLMs. Privacy-first architecture with multi-vendor adapter (OpenAI, Claude, Grok, Gemini, Qwen, DeepSeek). FastAPI + React + Telegram bot.
telegram_to_pdfVectorDB - Telegram chat export to AI-ready PDF converter. Smart chunking, dynamic sizing, optimized for vector databases and n8n workflows.
Investigating uncertainty quantification for autonomous web-browsing AI agents. Developed GUAWA - a framework implementing 5 uncertainty estimation methods based on 6 research papers.
Implemented methods:
| Method | Type | Based on |
|---|---|---|
| Normalized Entropy | Single-step | H(Y)/log(N) normalization |
| Predictive Entropy | Single-step | Information-theoretic H(Y) |
| Semantic Entropy | Single-step | NLI-based clustering |
| Averaging (RMS/mean/geometric) | Multi-step | SAUP paper |
| UProp (IU+EU decomposition) | Multi-step | UProp paper |
Core question: How can agents reliably self-assess confidence before executing real-world actions?
Integrated with REAL benchmark (AGISDK), browser-agent, and tree-search agent implementations. Hyperparameter calibration via Optuna.
Beyond technical execution, I invest in the human side of engineering and leadership:
- Public Speaking - Presenting technical solutions and product vision to stakeholders and at events
- Strategic Decision-Making - Cognitive frameworks for complex technical and business decisions
- Team Building & Communication - Growing a team from scratch, mentoring engineers in AI/ML and system design
- Business Modeling - Translating technical capabilities into viable business models and product roadmaps
- Emotional Intelligence - Managing cross-functional teams, client relationships, and high-pressure delivery cycles
- Research Methodology - Bridging academic research and production engineering, paper-to-product pipeline
Developed through ITMO University programs: public speaking, cognitive decision-making methods, business modeling, communications and team building, life in science.
|
Deep Learning — CNNs, Transformers, autoencoders, diffusion models
Orchestration — 50+ node LangGraph pipelines, conditional routing, fan-out/fan-in, HITL Domain-Driven — bounded contexts, aggregates, domain events, anti-corruption layers |
Backend — FastAPI (async, DI, middleware), SQLAlchemy, asyncpg, Alembic, Celery
Containers — 25+ service Docker Compose, multi-stage builds, health checks
Framework — Flutter (Riverpod, BLoC, GoRouter, Freezed, build_runner) |
| Year | Event | Result |
|---|---|---|
| 2025 | AI Talent Hub ITMO | 1st place |
| 2025 | Leaders of Digital Transformation | 3rd place |
| 2024 | Digital Almaty Product Hackathon | 2nd place |
| 2024 | AI Talent Hub ITMO | 4th place |
| 2023 | Kaspersky Hackathon | 1st place |
| 2022 | NASA Space Apps Challenge | 3rd place |
ITMO University - Faculty of Programming and Computer Technologies
Technical focus: LLM engineering, AI agent architectures, MLOps, distributed systems.
Leadership programs: Public speaking, cognitive decision-making methods, business modeling, communications & team building, scientific methodology.
Mentoring team engineers in AI/ML technologies and system design practices.
Mountain hiking, strategic board games, chess, world history and historical landmarks, travel. I read scientific papers for fun and play Civilization for the decision trees.
Open to technical advisory, architecture consulting, and strategic partnerships worldwide.