Vitaliy Kuzmin

Lead AI Architect & Senior Engineer

CV	Email	LinkedIn	Location	Open to
PDF	hello@hi-ai.app	Vitaliy Kuzmin	Tbilisi, Georgia	Remote / Full-time / Contract

Summary

Lead AI Architect and Senior Engineer with 9+ years of experience building complex software systems and 3+ years specializing in multi-agent orchestration, local LLM infrastructure, RAG, LLM evaluation, and private/on-premise AI. Proven track record of transitioning traditional enterprise workflows into AI-driven ecosystems. Founder of hAi, a local-first AI assistant with a public/open-source release planned. Deeply focused on AI agents, local models, LangGraph, MCP, and practical assistant platforms for real business systems.

Selected Projects

Axioma AI — enterprise AI assistant platform for asset management workflows.
hAi — personal local-first desktop AI assistant and R&D environment.
Smart Publish — AI-assisted publishing platform.
Urban Lighting — ML reconstruction platform for urban lighting: field inventory and mobile laboratory measurements turned into geospatial datasets, LightGBM-based illuminance prediction, and reconstruction planning.
Light in Night — professional lighting design software; team leadership for a 10+ person engineering team.

Experience

Lead AI Architect
InterProCom | Mar 2025 - Present

Leading the development of Axioma AI, an enterprise AI assistant platform for the Axioma asset management system used in the energy domain.
Multi-Agent Orchestration: Architected a multi-agent system using LangGraph, integrating specialized agents for SQL, norm-control, and domain RAG pipelines.
Private AI Infrastructure: Deployed and optimized local LLM infrastructure on company servers, enabling private AI usage without sending sensitive enterprise data to external services.
Evaluation & Observability: Designed LLM-as-a-judge and automated evaluation workflows with Langfuse for agent quality control.
AI Platform Engineering: Built the solution around open-source AI infrastructure, model routing, observability, CI/CD, Docker-based deployment, and enterprise API integrations.
Context-Driven Development: Organized architecture docs, global prompts, agent instructions, reusable skills, and project knowledge alongside the codebase to support AI-assisted engineering.
Business Impact: Reduced manual navigation across enterprise data and documentation by up to 50% in target workflows and expanded AI-assisted analytics and planning with a broader set of enterprise data sources.
Tech Stack: Python, FastAPI, LangGraph, LangChain, LlamaIndex, local LLMs, Ollama, LiteLLM, Langfuse, LLM-as-a-judge, RAG, SQL agents, PostgreSQL, Docker.

Founder & Lead AI Developer
hAi, local-first AI assistant | 2024 - Present

Designed and developed a personal local-first AI assistant focused on local execution, contextual memory, voice interaction, and tool use, with a public/open-source release planned.
Core Engine: Built a distributed local application using Python/FastAPI, agent orchestration, LiveKit for voice interactions, and ActivityWatch for computer activity context.
MCP Integration: Implemented Model Context Protocol support to allow agents to interact with local tools, files, and external systems.
Local LLM R&D: Use hAi as a practical environment for Mac Studio inference optimization, local models, AI coding assistants, memory systems, and agent UX.
Tech Stack: Python, FastAPI, LangGraph, LangChain, LiveKit, React, Electron, SQLite, LanceDB, ActivityWatch, MCP, Ollama, Langfuse, GCP.

Lead AI Engineer
BL Group R&D Center | Jul 2022 - Mar 2025

Led the development of Smart Publish, an AI-assisted publishing platform, moving the idea from an early prototype toward a working LLM-powered product.
Architected an automated publishing workflow for raw news drafts from multiple channels, including email and Telegram.
Designed the LLM processing pipeline for structuring, rewriting, editing, formatting, and preparing publication metadata, helping prepare multi-channel publications from a single draft.
Integrated the platform with a client CMS through an API and prepared the workflow for social media distribution.
Business Impact: Reduced repetitive editorial preparation and formatting work by up to 60% and improved content consistency through channel-specific adaptation for websites and social platforms.
Tech Stack: Python, FastAPI, OpenAI API, Telegram API, REST APIs, CMS integrations, Google Cloud Platform (GCP), n8n, prompt engineering.

Team Lead / Lead Software Engineer
BL Group | Apr 2017 - Jun 2022

Grew from Full Stack Developer to Team Lead, managing cross-functional teams of up to 10 engineers across C++, Python, frontend, QA, design, and analytics.
ML Reconstruction Platform: Led the development of Urban Lighting, making field data reusable across reconstruction planning projects.
Enterprise Automation: Automated field data collection, lighting asset mapping, reconstruction analytics, and tender documentation workflows.
Product Development: Led development processes for Light in Night, scaling the engineering team to 10+ people and improving release predictability.
Business Impact: Accelerated field data collection and processing by up to 60%, reduced repeat city survey visits by up to 70%, and removed dependency on an external map-markup contractor.
Tech Stack: Python, Node.js, C++, Docker, PostgreSQL, LightGBM, GitLab CI/CD, React/Vue.

Skills

AI & LLM: LangGraph, LangChain, LlamaIndex, RAG, SQL agents, multi-agent systems, MCP, local LLMs, Ollama, LiteLLM, LiveKit, prompt engineering, Langfuse, LLM-as-a-judge.
Backend & Architecture: Python, FastAPI, system design, microservices, REST, gRPC, Docker, Google Cloud Platform (GCP), PostgreSQL, Redis.
Infrastructure & Data: private/on-premise AI deployment, local model optimization, CI/CD, DevOps, vector databases, LanceDB, FastEmbed, OpenTelemetry.
Leadership: team leadership, hiring, mentoring, Agile/Scrum/Kanban, product architecture, roadmap planning.

Education

Moscow Power Engineering Institute (MPEI) — Master’s Degree, Electronics and Nanoelectronics, 2018