Русский · English
AI Product Lead — designing, building, and shipping LLM-driven products end-to-end.
13 years in IT. Started in network and systems administration, moved through project management, and the last few years went deep into LLM engineering and AI product work. Comfortable owning architecture, code, and product decisions on the same team — the kind of role where the same person picks the model, designs the agent flow, writes the API, and watches the production logs.
📫 Telegram @Crimson_Beherit
agentic-workflow — multi-agent framework for Claude Code
Tier-aware acceptance (S/M/L), filesystem-isolated adversary review in fresh subprocesses, cross-family second opinion via Codex MCP (GPT-5), human as supreme judge at critical transitions. 60 specialized agents, 45 methodology skills, 21 Python orchestration scripts.
Solves three failure modes of single-model agent pipelines: framing contamination, Goodhart on validators, and undifferentiated rigour.
Stack: Python · Claude · Codex · MCP · Mermaid
call-analytics-system — LLM call analytics service (architecture case-study)
Production LLM service for sales / customer-service call analysis. Ingests calls from CRMs, runs STT + diarization + LLM scoring, pushes results back. SaaS with on-premise readiness via provider abstractions. End-to-end ownership: architecture, backend, frontend, deployment.
Stack: Python · FastAPI · async SQLAlchemy · Celery · Redis · PostgreSQL · React · TypeScript · Fernet encryption · JWT + RBAC
content-generation-pipeline — composable content generation platform (architecture case-study)
Multi-tenant platform with a UI pipeline builder, reusable stages, human-in-the-loop approvals as a first-class stage type, and multi-channel publication scheduling. Solo full-stack ownership from product idea to operations.
Stack: Python · FastAPI · async SQLAlchemy · Celery · Redis · PostgreSQL · React · TypeScript · Fernet encryption · JWT
hh-bot — Telegram bot for LLM-driven job filtering
Browser-automation pipeline for vacancy search, LLM-based relevance scoring with score-driven routing (auto-apply / manual / skip), cover letter generation, autopilot loop with Telegram delivery.
Stack: Python · aiogram · Playwright · SQLite · OpenAI-compatible LLM client
AI & LLM engineering Designing agentic workflows, picking models for the job, prompt engineering, RAG, evaluation. Comfortable across major LLM providers — Anthropic, OpenAI, Google — plus open-weight options for cost-sensitive cases, and the differences that matter day-to-day: context windows, tool-use semantics, output reliability, latency. Two patterns I keep coming back to — Individual AI (personal- productivity skills and agents for one employee) and Institutional AI (agents wired into team and product processes).
Backend Python is my main language. Async web services and APIs, task queues and background workers, ORMs, message brokers, relational and key- value storage. Comfortable with the security fundamentals you need to take a service to production: auth, encryption at rest, rate limiting, input validation, the OWASP top-10 surface.
Frontend React + TypeScript ecosystem for internal tools, admin panels, dashboards, and product UIs. Component libraries, design tokens, charts. Not a frontend specialist by trade — but enough to ship a product when there isn't a dedicated frontend developer on the team.
Operations & deploy Docker, Linux administration (Ubuntu / CentOS / Debian), Git workflows, CI/CD pipelines (GitLab CI, GitHub Actions). Comfortable taking a clean VPS to a running production server: nginx, systemd, TLS, monitoring, alerting. Have done this on bare-metal and on managed platforms.
Product & project management PM since 2018, including running up to 25 parallel client projects in a web-development practice. Methodology picked per project — Scrum, Kanban, LeSS, Waterfall hybrids — instead of forcing one template. Discovery, KPI definition, stakeholder management, scope and budget control, hiring and onboarding for small teams.
Infrastructure roots (earlier career) Spent several years as a sysadmin, then leading an IT department, before moving into PM and AI work. Networking (MikroTik / Cisco), Active Directory, Windows Server, Linux server admin, telephony (Asterisk), monitoring. Useful background for end-to-end AI products — when something breaks below the application layer, I can go look.
AI Product Lead · AI Implementation Lead · AI Product Manager · solo or lead engineer roles on small AI-product teams.