Forward Deployed AI Engineering · Embedded engagements via Toptal
Your AI pilot, in production. Not in a slide deck.
88% of enterprise AI pilots never reach production. I'm the engineer who makes yours one of the 12%.
Source: IDC × Lenovo CIO Playbook 2025
Already worked with
Selected work
AI systems shipped into production.
OneLiner — Multi-Agent Event Pipeline
In ProductionProduction multi-agent system extracting structured intelligence from raw event streams. End-to-end pipeline: ingestion → multi-agent extraction → automated reporting. Multi-LLM routing for cost optimization.
Multi-Agent · LLM Orchestration · Production Pipeline
FirmPilot — Legal Data Engineering for LLM Knowledge Base
In ProductionBuilt the data preparation layer that feeds FirmPilot's legal AI: nationwide US court-opinion ingestion across 10+ jurisdictions, PDF→Markdown conversion at scale, S3-backed coverage reports. Production data engineering for downstream LLM systems.
AI Data Engineering · Web Scraping at Scale · LLM-Ready Knowledge
Majesti AI — Own Production AI Playbook
Live · Solo-builtLive Telegram product. Multi-LLM routing across 4 providers (Gemini / ByteDance / Topaz / Suno) via MCP architecture. Semantic classification through prompt-based routing — 1965-record eval dataset. Custom voice/vision pipelines. Same playbook I deploy for clients.
Multi-LLM · MCP Architecture · Prompt Routing
Mysleformy — Multi-Tenant Agent Marketplace (2023)
Pre-OpenAI Assistants StoreMulti-tenant agent platform with semantic search for discovery: users create LLM agents via system prompts, agents indexed in Qdrant vector DB, semantic search finds relevant agents by query. Public/private visibility. Same pattern as today's GPT Store — shipped before OpenAI launched theirs. Market need shifted when OpenAI released their own.
Production RAG · Vector DB · Multi-Tenant AI Architecture
Three ways to work together
All engagements run through Toptal — same hourly rate, same direct access to me.
AI Deployment Sprint
In 2–6 weeks I take your AI pilot to production. Architecture, infrastructure, observability, cost controls — everything blocking the release.
When: You have a POC working in Jupyter, but it never makes it to production.
Work with me on this →Multi-Agent System Build
End-to-end multi-agent pipeline: orchestration, tool use, state management, fallbacks. I work across stacks — direct APIs, LangGraph, MCP, Anthropic Agents — choosing what fits the task, not imposing a framework.
When: You need a production multi-agent system, not a toy.
Work with me on this →Production AI Audit
Review of your existing AI project. I find bottlenecks: cost, latency, accuracy, prompt fragility. You get a prioritized fix plan.
When: Your AI product is running, but "something is off".
Work with me on this →From sketch to scaled product
First two stages — validate yourself with no-code. From stage 3, I architect the MVP that scales, and lead you to a revenue engine.
Spark
AloneManual experiment with 10 customers, Excel + messenger
Validate
+ no-codeNo-code automation (Zapier + Bubble), 100 customers
Build
+ AlexCustom MVP, own database, 1000 users
Scale
+ Alex + DevOps + GrowthProduction-grade platform, dedicated team, ARR tracking
Refine
+ Alex + Full teamFully automated growth engine, self-onboarding, viral loops
17 years engineering. Forward deployed since 2021.
17 years engineering. 11 years Toptal — Top 3% senior talent. R&D from particle accelerators (CERN, BINP) to AI startups (FirmPilot, OneLiner).
Now focused on Forward Deployed AI Engineering: helping teams ship AI pilots into production. Work across stacks (direct APIs, LangGraph, MCP) without framework cults — choosing what fits the task.
Open-source: @alexmakeev/llmems — long-term memory module for LLM agents (Zettelkasten + Matryoshka embeddings, PostgreSQL+pgvector).
Active R&D: Mysterra — multiplayer game with LLM-as-game-engine on top of this memory module.
Read full bio →Where I go deep
Three verticals, one engineer. Each backed by shipped projects.
AI-Native Engineering
I design multi-agent systems and LLM orchestration that turn unstructured input into production-ready knowledge. For FirmPilot, I built the data preparation pipeline (scrapers across 10+ jurisdictions, PDF→Markdown, coverage reports) that feeds their legal AI. My own products run the same playbook — Majesti AI ships features end-to-end through Claude-driven development.
Innovation R&D Leadership
I lead small R&D teams toward business outcomes — not just code, but the first working version of something new. Founder-engineer at OneLiner (Heroku-style PaaS on Kubernetes), tech lead at FirmPilot. Intel's School of Leaders for Innovative IT Projects (NSU, Novosibirsk, 2009) trained me in this discipline early.
Tech ↔ Business Bridge
I translate between business stakeholders and engineers. Years embedded in management consulting (ASIA Consulting Group, OneBox — acquired by EY) taught me the language of executives: frameworks, diagnostics, board-ready output. Presidential Management Program added formal training in finance and organizational strategy. That bridge is rare in senior engineers and is where I'm sharpest.
From CERN to AI startups
Since 2007. Three eras. Accelerating velocity.
Science
2009–2014
CERN, Budker INP, Novosibirsk State University. Accelerator software, image processing, teaching the next generation of engineers.
Platform
2014–2022
Toptal Top 3%. Built lead-generation engines that drove millions in revenue. Scaled web platforms across industries.
AI-Native
2021–now
Multi-agent systems, LLM cost optimization, production AI infrastructure. From GPT release date to today's frontier.
Ready to ship your AI pilot?
I take 1–2 new engagements per quarter. All through Toptal — they handle contracts, payments, and legal. You work directly with me.