Alexey Meshkov

I take LLM products to production

AI Product Manager. Founder of Side, an AI service that analyses legal documents. Before that, 15 years running sales and operations, most recently in e‑commerce and on marketplaces.

plantsherbivorespredators what is this world

Side: AI analysis of legal documents

A person uploads a court document or a claim; Side analyses it and drafts a response. In production since 15 June 2026. Built solo, from discovery and architecture to evals and distribution.

The pipeline runs from cheap steps to expensive ones, so each next step gets less work:

  1. OCR and parsing. PDFs and scans into structured text.
  2. Pseudonymisation. Four stages, personal data recall of 95%+. No name leaves the system — a requirement of Russia’s personal data law (152‑FZ).
  3. Classification on Claude Haiku. Identify the type of situation and filter out what the service must not help with.
  4. Analysis on Claude Sonnet. Assessment of the position and a draft document.
  5. Verification. An LLM judge and a stop classifier with target error rates: at most 1% misses, at most 15% false alarms.
70 test caseseval pipeline with an adversarial generator and a red‑team gate before release
3 regionsRussia, the Netherlands, the US: citizens’ data never leaves the Russian region
3 modelsbenchmarked: Claude Haiku, YandexGPT 5.1 Pro, Kimi K2.6

Portfolio

ecom-ai: AI logistics for marketplaces

R&D and consulting for large sellers and warehouse logistics: sales forecasting, stock, returns, allocation across warehouses.

mp-ai: AI store audit

An audit of a store on Wildberries, Ozon and Yandex Market for small businesses: where listings and prices lose money and what to fix first.

sorting-ai: AI warehouse sorting

Automating warehouse sorting: fewer errors, faster sorting, and flows modelled in simulation before rollout. An R&D prototype built with Python, PyTorch, Blender (synthetic scenes) and SimPy (flow simulation).

ecom-ai and mp-ai used to live on separate domains. Now they are here; together they had about 2,300 organic visitors a month. sorting‑ai is an R&D prototype and never had a public site.

Career

  1. Founder, AI Product Manager — Side, ecom-ai, mp-ai

    The full cycle of my own AI products, solo.

  2. Commercial Director — AI Technologies LLC

    Account management and analytics services for marketplace sellers. Clients’ combined turnover above 3.5 billion roubles a month.

  3. Head of Marketplaces — Bi&Ki, children’s footwear

    Built the channel from zero to 38 million roubles a month in 9 months. Team of 6.

  4. Operations and Commercial Lead — apparel store on Ozon and Wildberries

    Assortment, sourcing in China and Turkey, helping foreign companies enter Russian marketplaces.

  5. Executive and Operations Director — Chisty Dom, Torgovy Dom 8

    Public procurement, 200+ suppliers, store openings together with X5 Retail Group.

  6. Project Manager — Nikpastroy, Genezis IS

    Infrastructure: MEGA shopping centres, Serebryany Bor tunnels, Russian Railways.

Education: Lomonosov Moscow State University, Mechanics and Mathematics (2004); Russian Foreign Trade Academy, international finance (2012).

Lab: when does evolution pay for a brain

Why nature needs a brain, and what that can teach AI. An essay on intelligence as a response to the environment, with the brain as only one way to respond. Perhaps AI needs not a brain, but a world in which a mind becomes necessary.

Iteration 1: a 2D world where a neuron costs energy. I tested whether evolution grows a brain when the climate changes faster. Short answer: no — without memory and learning the brain has nothing to do.

Iteration 2: when does evolution pay for learning. I gave the animals lifetime learning. It evolves where the world changes between generations but stays stable within a lifetime. The brain still did not grow.

Can the model reproduce the lynx–hare cycle. I checked the ecosystem against 90 years of real Hudson’s Bay Company data. The period is easy to match; the lag and amplitude are not. An honest look at where the model breaks.

Work and consulting

I am looking for an AI Product Manager role in a team that ships LLM products to production. Strongest where the domain is e‑commerce, marketplaces and logistics.

I consult on adopting LLMs: model choice and inference cost, evals without labelled data, handling personal data under strict privacy law.