Особисті дані приховані

Mykhailo Melnyk
AI engineer
Lviv, Ukraine | [відкрити контакти](див. вище в блоці «контактна інформація») | [відкрити контакти](див. вище в блоці «контактна інформація») | LinkedIn | GitHub

AI Engineer with 6 years of programming experience – from early Telegram bots, data automation, and market monitoring
tools to classical ML, scientific computing, and computer vision – now focused on RAG and agentic systems. Open to ML/AI
engineering roles
Projects
Crypto Agent LangGraph, FastAPI, React, MLflow, Tavily, DuckDB
AI assistant automating crypto analysis across recent moves, funding and news
● Used SQL queries over cached Binance REST API calls as primary analysis tool with DuckDB sandboxing and query
wrapping for security and hard row limit for token economy
● Added Tavily web search – isolated as untrusted input to guard against prompt injection – with news vs general search
differentiation, persistent user memory, and reusable skills for recurring requests
● Shipped as a React/FastAPI web app with query logging for understanding user needs and MLflow tracing for debugging
Hierarchical RAG system LangChain, LangGraph, Ragas, ZenML, MLflow, MongoDB, Qdrant
RAG system that leverages a knowledge base's natural hierarchical structure to guide retrieval, letting it answer both factual
and structural questions
● Designed a routed RAG pipeline that first decides the appropriate retrieval method and then uses query rewriting, cross-
encoder reranking and chunk location metadata improving the search
● Built an agentic alternative that lets the model freely traverse the hierarchy itself with size-constrained expansion
● Built the data pipelines around change-data-capture and versioned configs, using ZenML for pipeline monitoring and
MLflow for experiment tracking
● Designed a custom LLM-as-judge answer-correctness metric (NLI entailment against atomic facts); improved answer
correctness from a 0.50 naive-chunking baseline to 0.737 (hierarchical RAG) and 0.755 (agentic) on an early eval set
Credit risk modelling system Python, scikit-learn, optuna, FastAPI, Django, Docker Compose, LightGBM, XGBoost
● Compared regression and classification methods for threshold-optimized loan approval across 151 raw features, each
preprocessed with feature-specific missingness and encoding strategies, achieving 1.73x aggregated return improvement
over an approve-everyone baseline on holdout set using a LightGBM regression model
● Deployed as FastAPI inference service and a Django web application – orchestrated with Docker Compose
Experience
Aerostat prediction | Python, xarray, Qt, Leaflet.js, SQLite September 2024 – October 2025
Client-commissioned novel scientific application for high-altitude balloon path forecasting
● Owned the full product lifecycle across specification, research, algorithm design, architecture, implementation and
testing, expanding a short initial client brief into the full scope delivered
● Gave users the flexibility to choose from a wide range of data sources (GFS, radiosonde measurements, national and
regional models) and rank them – using the most accurate source given its coverage and simulated balloon position
● Added the ability to pre-download the data for offline use and use third-party grib/netcdf files through template adapters
● Implemented multi-path rendering on an interactive map, letting users compare predictions across different data sources,
forecast times, and actual flown paths – with project save/load support
● Validated predictions against real balloon launches, presenting findings at SoftTech-2025
Other projects
Logo detection system – fine-tuned YOLOv8 on a custom 1,091-image annotated dataset for real-time video detection
Petri net simulator – modeled a bank's process as a Petri net to analyze productivity and the effect of structural changes
Wind climatology model – used radiosonde historical data for year-ahead launch planning; handled missing values through
linear, analogue, and neighboring-station imputation; found no signal beyond climatology over lagged features
Skills & Certifications
Development: Python, SQL, JavaScript, FastAPI, Django, Git, Qt
Infrastructure: Docker, Docker Compose, MongoDB, SQLite, ZenML, MLflow
AI engineering: LangChain, LangGraph, RAG, agentic systems, Ragas, Qdrant, Tavily
Machine Learning: scikit-learn, pandas, xarray, matplotlib, plotly, Optuna, Ultralytics, Albumentations
Languages: English (professional working proficiency), Ukrainian (native)
Education
Kyiv Polytechnic Institute | Bachelor’s degree Kyiv, Ukraine | September 2021 – June 2025
Information systems software engineering
Relevant coursework: Neural Networks, Data Science, Information Systems Infrastructure (AWS, Azure, VMware labs)