APPLIED AI · AI ENGINEERING · PRODUCTION
Design and ship
production AI systems.
My name is Vadim Samoilov. I’m an Applied AI / ML Engineer and a graduate of Saint Petersburg Electrotechnical University “LETI”. I have about 1.5 years of commercial experience building AI services and agentic systems end to end — from requirements and architecture to backend, AI logic, integrations, infrastructure, production release, and stakeholder presentation.
LLM · AI Agents · RAG · MCP
Python · Backend · Integrations
Full‑cycle delivery
BACKGROUND · EDUCATION · PATH TO AI
From engineering foundations to production AI.
My path into AI developed step by step: LETI gave me a foundation in mathematics, programming, and business processes; an independent gamedev project gave me hands-on engineering and team experience; logistics and automation taught me how software connects to real operations. Since 2023 I have focused systematically on Python and ML, and since 2025 I have been building commercial AI systems.
2013 — 2017 · Bachelor’s degree
SPbETU “LETI” · Innovation Management
An interdisciplinary program spanning business, management, and technology. Coursework included Pascal, Python, C++, neural-network fundamentals, calculus, mathematical economics and modeling, physics, and business English.
IEEE publication on warehouse logistics ↗012013—2017
University
Mathematics, programming, and business processes
The program combined several disciplines — from programming and neural-network fundamentals to mathematical modeling, economics, and management. Together with my logistics lecturer, I co-authored an IEEE paper on innovative management methods in warehouse logistics.
022015—2018
Hands-on engineering
My first substantial independent software project
Alongside university, I taught myself game development and released the “Wind of Time” modification for S.T.A.L.K.E.R. I worked with C++, Lua, XML, X-Ray Engine, the game SDK, and graphics tools, then assembled a small team to continue developing the project.
032018—2022
Logistics · automation
From business operations to algorithms
I worked in warehouse logistics for a major automotive dealer in Saint Petersburg. In parallel, I explored algorithmic trading and AI for time series and built a prototype automated trading bot with an experienced trader. The experiment did not produce a financial result, but it brought me back to systematic software development.
042023—2024
Python · Machine Learning
A systematic move into ML engineering
I studied Python and ML in depth through Stepik, karpov.courses, and independent projects — progressing from linear regression, gradient boosting, and classic NLP to multilayer neural networks, Transformers, and LLMs, while also taking part in ML competitions and team projects.
052025—present
Commercial AI
Full-cycle production delivery
At Imprice and OWEN-Energo Group, I moved into commercial AI engineering. Today I build systems almost from scratch: clarify requirements, design architecture and integrations, implement backend and AI logic, set up infrastructure, ship to production, and present the result to stakeholders.
FEATURED CASE STUDIES
Production case studies
Two projects that best demonstrate my direct contribution, system-level architecture, and delivery into real-world use.
01 / COMMERCE AI
RGM BASE · AI EQUIPMENT SELECTOR
Equipment selection across 170,000+ SKUs and 14+ brands
I independently designed and developed the backend, AI logic, core UI, and integrations for the current RGM Base AI equipment selector. The system works with 170,000+ SKUs across 14+ brands; the MCP architecture is designed to scale to dozens of brands and more than one million products. Background and asynchronous tasks run on RabbitMQ + Celery. The previous production version for three brands handled real customer requests, selected equipment, and handed conversations over to a human operator when needed.
Current catalog170 000+SKUs · 14+ brands
Previous version3 brands≈ 3–4% qualified leads / month
Architecture1M+target product scale
02 / HR AUTOMATION
AI INTERVIEWER IN HH.RU
Two-stage automated interviews without third-party messengers
I designed and developed a two-stage AI interviewer that communicates with candidates directly in hh.ru chat, automatically creates and updates candidate records in Bitrix24, attaches CVs, moves applicants through hiring stages, and supports take-home assignments. Queues and background tasks are implemented with RabbitMQ + Celery. The system is used in real hiring, and candidates who passed through it have already been hired.
Interview2 stagesautomated screening
Channelhh.rudirectly in platform chat
Real-world useSuccessful hirescandidates have already been hired
PROJECT EXPLORER
Projects by category.
Production and commercial systems come first; competitions and pet / hobby work are separated so they do not blur the professional experience.
ML COMPETITIONS
Competitions are a distinct part of my ML experience.
I have competed both in teams and individually across NLP, classification, vector search, multimodal RAG, labeling interfaces, and resource-allocation algorithms.
🥉3rd placeJarvis Labeling · ecom.tech
044th placeecom.tech × Deep Learning Lab
FINFinalistIT Purple Hack 2025
01
Jarvis LabelingSupabase · Streamlit · Pandas · data labeling
3rd place ↗
02ecom.tech × Deep Learning LabBERT · multilabel classification · Streamlit
4 / ~100 ↗
03IT Purple Hack 2025RAG assistant + VM placement / First‑Fit Decreasing
Finalist ↗
04Multimodal RAG: ColPali + Vikhr-2-VLByaldi · ColPali · Vikhr-2-VL · PyTorch · Transformers · Streamlit
GitHub ↗
EXPERIENCE
Applied AI: models, backend, integrations, and delivery.
My strength is not a single model or API, but assembling complete systems: requirements, architecture, Python backend, AI logic, queues, databases, integrations, infrastructure, and production delivery.
requirements→architecture→code→infra→production
08.2025 — present
OWEN-Energo Group
Junior ML Engineer
AI services, agentic systems, LLMs, integrations, and infrastructure within the company’s IT department.
- Built the development and deployment infrastructure for AI projects from scratch.
- Designed integrations with Bitrix24, hh.ru, OData/1C, and other APIs.
- Shipped production systems for sales, HR, recommendations, search, and voice AI.
05.2025 — 07.2025
Imprice
Junior AI Developer
AI-agent engineering team: Python wrappers, microservices, testing, RAG pipelines, and databases.
- Built tools for integrating agents into real workflows.
- Built RAG search over CRM documentation.
- Monitored Telegram chats: GPT-based topic segmentation, completion/outcome/sentiment classification, and Google Sheets metrics refreshed every 30 minutes.
TECH STACK
Core technology stack.
Grouped by engineering function rather than as a keyword dump: AI/ML, backend, data/infra, and automation.
vLLM + llama.cpp
Running local models and integrating them into production services.
n8n + Dify
Production automations, search bots, and recommendation systems.
Dagster + OpenSearch + S3
Data pipelines, object storage, and search infrastructure.
Architecture to release
Architecture, development, infrastructure, and stakeholder presentation.
PET PROJECT · WINNER
MoodMeter
A team-built sentiment analysis system using Python, Hugging Face Transformers, PostgreSQL, and Streamlit. The project won the internal competition in the karpov.courses ML simulator.
TransformersPostgreSQLStreamlitNLP
Project GitHub ↗
HOBBY ENGINEERING · GAME DEVELOPMENT
“Wind of Time” · S.T.A.L.K.E.R.
A game modification released in 2017. I worked not only on content but also on the technical side of the original X-Ray Engine and game systems. The project remains visible in the community and ranks top-2 by number of reviews among hundreds of projects on a major fan site.
- C++ · Microsoft Visual Studio: edited X-Ray source code across all three original S.T.A.L.K.E.R. games.
- Lua: wrote and refactored scripts, character behavior, and gameplay logic.
- XML: configs, character parameters, quests, and gameplay logic.
- Game SDK: level/game design and world building; modeled selected assets in Autodesk 3ds Max.
- Graphics: created, edited, and adapted in-game textures in Paint.NET and Adobe Photoshop.
C++LuaXMLX‑Ray EngineGame SDK3ds MaxPhotoshop
Project page ↗
NEXT ROLE
Looking for a role where AI becomes part of a real product.
Telegram is the fastest way to reach me.