← Back to list
Senior
Registration: 10.11.2022

Maksim Mezhov

Specialization: AI Engineer / Data Scientist
— I design and implement AI systems that solve practical problems — from industrial process automation to AI-assisted user interfaces. — My focus is on building solutions that actually work in production and deliver real value. — With 5+ years of experience in AI and Data Science, I’ve worked across two key domains: — Industrial AI — failure detection, fuzzy control systems, process optimization. — LLM-based applications — agentic systems, retrieval-augmented generation (RAG), multimodal assistants. Recent projects include: — A multimodal RAG system for personalized fashion recommendations. — A plant documentation assistant powered by locally hosted LLMs and custom embeddings. — ViGist – a tool that extracts key insights from long-form YouTube videos. — AI agents for automating interactions in education and services. — Winner of DigitalSkills-2021 in Machine Learning & Big Data. — Co-author of 3 research publications in AI. Publications: — Digital Skills 2021. — Using machine learning to determine accounting codes based on the economic rationale of procurement documentation. — Anomaly Detection in a Granulation Plant for Uranium Dioxide Powder Pressing into Fuel Pellets. — Machine Learning Model for Early Detection of COVID-19 by Heart Rhythm Abnormalities.
Data Scientists Python Microsoft SQL server Django Java MySQL scikit-learn Tensorflow Software Development FastAPI Process automation Neural networks NumPy LLM Qdrant Machine Learning Deep Learning Algorithms RAG Artificial Intelligence
— I design and implement AI systems that solve practical problems — from industrial process automation to AI-assisted user interfaces. — My focus is on building solutions that actually work in production and deliver real value. — With 5+ years of experience in AI and Data Science, I’ve worked across two key domains: — Industrial AI — failure detection, fuzzy control systems, process optimization. — LLM-based applications — agentic systems, retrieval-augmented generation (RAG), multimodal assistants. Recent projects include: — A multimodal RAG system for personalized fashion recommendations. — A plant documentation assistant powered by locally hosted LLMs and custom embeddings. — ViGist – a tool that extracts key insights from long-form YouTube videos. — AI agents for automating interactions in education and services. — Winner of DigitalSkills-2021 in Machine Learning & Big Data. — Co-author of 3 research publications in AI. Publications: — Digital Skills 2021. — Using machine learning to determine accounting codes based on the economic rationale of procurement documentation. — Anomaly Detection in a Granulation Plant for Uranium Dioxide Powder Pressing into Fuel Pellets. — Machine Learning Model for Early Detection of COVID-19 by Heart Rhythm Abnormalities.
Data Scientists Python Microsoft SQL server Django Java MySQL scikit-learn Tensorflow Software Development FastAPI Process automation Neural networks NumPy LLM Qdrant Machine Learning Deep Learning Algorithms RAG Artificial Intelligence

Portfolio

MSZ (Elemash)

● Lead of Document Management team.

Tsifrum (Cifrum)

● Reduced workflow time of the conformity assessment documents for products by 95% by creating a classification machine learning model. ● Automated a workflow process for detecting COVID-19 at an early stage using abnormalities in heart rhythm by creating a machine learning model which allows the detection of coronavirus in 83% of cases using data from mobile devices. ● Extended continuous operation period of a flame reactor by 30% by creating a machine learning model to anticipate clogging using time-series data from sensors. ● Sped up the procurement documentation classification process by 81% by developing a Streamlit-based Python web application using TF-IDF and classic machine learning algorithms.

Digital Technologies and Platforms LLC

● Developed and implemented a minimum viable product (MVP) of a recommender system, utilizing three machine learning models to achieve a 12% boost in phosphoric acid production. ● Completed an end-to-end Proof of Concept (PoC) for an industrial data science project aimed at reducing methanol production costs.

Skills

Python
Microsoft SQL Server
Django
Java
MySQL
MyStem
Scikit-Learn
TensorFlow
LSTM
Machine learning
Data analysis
PyTorch
Time-Series
C#
Industrial AI
Process Automation
Fuzzy Logic

Work experience

Data Scientist
since 04.2024 - Till the present day |Taiga Dynamics
Python, Fuzzy Logic, FastAPI, NumPy, Qdrant, Custom Embeddings, Multimodal LLMs, Parallel Computing, Time Series Analysis.
● Building AI-driven systems with a focus on industrial automation, real-time control, and LLM-based retrieval systems. ● Hands-on experience with RAG architectures, fuzzy logic, and multimodal LLMs. Key Achievements: ● Context-aware assistant for plant documentation — Designed and implemented a custom Retrieval-Augmented Generation system, manually building a ColPali-style embedding architecture for hierarchical chunking and context preservation. Integrated with a locally hosted multimodal LLM to enable accurate question answering over technical and visual documents. ● Failure detection for flotation & grinding circuits — Developed a rule-based fault detection system using historical telemetry and parallel processing. Delivered an 8.37 percentage point improvement in equipment availability, boosting metal recovery (Ni, Cu, Co) by 10% during trials. ● Real-time fuzzy control for SAG mill & hydrocyclones — Built fuzzy controllers to automate SAG mill speed and hydrocyclone feedwater regulation. Resulted in a ≥30% improvement in stability of critical process parameters, reducing operator workload and enhancing system robustness. ● Mentorship & internal tooling — Supported junior team members and introduced reusable development workflows to accelerate iteration and deployment.
AI Engineer
02.2024 - 04.2025 |AI Startup
LLMs, Agentic Systems, Multimodal RAG, Python, OpenAI API, Hugging Face, LangChain, FastAPI, Flask, Telegram API, PostgreSQL, Docker, Celery, Qdrant, Multimodal Embeddings, Agentic Design Patterns
● Building and deploying AI-driven products using LLMs, AI-agent frameworks, and workflow-based automation. ● Focus on fast prototyping, user-centric tooling, and production-grade system integration. Key Projects: ● Multimodal RAG-based stylist assistant — Developed a Retrieval-Augmented Generation system combining text–image embeddings to match user-uploaded outfit photos with stylistically relevant products from a marketplace. Built an LLM-powered assistant that interprets visual input and returns style-consistent recommendations with live product links. ● AI agent for beauty service booking — Implemented a tool-using AI agent (function calling) to handle customer dialogs, check CRM availability, and schedule appointments in real time. ● AI-powered assistant for a coding school — Created a multi-step agent pipeline with parallelized evaluators and optimizers to qualify leads, personalize answers, and enroll students automatically. ● ViGist – YouTube video summarization assistant — Designed an agent framework that captures key insights and highlights from long-form videos. ● AI content curator for the Data Science community — Built a chained prompt-based workflow that autonomously curates, formats, and publishes daily digests.
Data Scientist
07.2022 - 04.2024 |EuroChem
MVP, ML
● Developed and implemented a minimum viable product (MVP) of a recommender system, utilizing three machine learning models to achieve a 12% boost in phosphoric acid production. ● Completed an end-to-end Proof of Concept (PoC) for an industrial data science project aimed at reducing methanol production costs.
Data Scientist
04.2020 - 06.2022 |Rosatom Global
ML, Python, Algorithms
● Reduced workflow time of the conformity assessment documents for products by 95% by creating a classification machine learning model. ● Automated a workflow process for detecting COVID-19 at an early stage using abnormalities in heart rhythm by creating a machine learning model which allows the detection of coronavirus in 83% of cases using data from mobile devices. ● Extended continuous operation period of a flame reactor by 30% by creating a machine learning model to anticipate clogging using time-series data from sensors. ● Sped up the procurement documentation classification process by 81% by developing a Streamlit-based Python web application using TF-IDF and classic machine learning algorithms.

Educational background

Applied Mathematics and Computer Science (Masters Degree)
2004 - 2009
Ivanovo State Power Engineering University

Languages

EnglishUpper Intermediate