Artem Kholmansky
LLM Developer
Python
SQL
Invite to interview
Message
Download CVCV
Overview
Technical skills
Timeline
Roles
Overview
NLP/LLM Engineer with 3+ years of experience building LLM assistants using retrieval (RAG) and hybrid search, including LLM fine-tuning with PEFT methods and model deployment via vLLM, Kubernetes, and FastAPI. Also works as a DL engineer on training neural models for XPS spectrum decomposition and entity recognition, improving processing speed and reported metrics using neural architectures and distillation/ensembling.
Technical skills
Python• Middle • 3y+
SQL• Middle • 3y+
Python
FastAPI• 3y+
Databases
FAISS
Qdrant
PostgreSQL• 3y+
AI/ML
LoRA
Mistral
NER
Optuna
QLoRA
Qwen
Accelerate• 3y+
AWQ• 3y+
CatBoost• 3y+
ClearML• 3y+
GPTQ• 3y+
LangChain• 3y+
LLM• 3y+
NLP• 3y+
ONNX• 3y+
PEFT• 3y+
PyTorch• 3y+
RAG• 3y+
Scikit-learn• 3y+
Transformers• 3y+
vLLM• 3y+
NumPy
PyTorch Lightning
SciPy
DevOps
Rest API
Docker• 3y+
Kubernetes• 3y+
Git
Timeline
Scientific Researcher / DL Engineer
•
Middle
IK SO RAN
•
Full-Time
Trained neural models for automatic decomposition of XPS spectra used by researchers across multiple institutes, reducing average decomposition time. Built an XPS preprocessing and training pipeline, and trained encoder-only entity recognition models with BIO labeling, achieving F1 as reported. Modified and trained recurrent and convolutional networks (biLSTM, biGRU, and 1D U-Net with self-attention) and applied ensembling and knowledge distillation to speed up batch processing while keeping comparable F1. Exported/handled models for production workflows using ONNX and managed experiments with ClearML.
Python
PyTorch
PyTorch Lightning
Transformers
Scikit-learn
NumPy
SciPy
ONNX
Git
ClearML
NLP/LLM Engineer
•
Middle
Elektrokhimzashchita LLC
•
Full-Time
Developed an LLM-based assistant with retrieval to automate internal company processes and support a large share of employees. Built a hybrid dense/BM25 retrieval pipeline using Qdrant and PostgreSQL, including GraphRAG-style retrieval, HyDE, parent-child retrieval, and reranking to improve document search quality. Fine-tuned and deployed LLMs for company-specific tasks using SFT/DPO/SimPO/ORPO with PEFT methods, Accelerate and FSDP, plus GPTQ/AWQ quantization; improved generation quality by an internal metric. Deployed serving with vLLM on Kubernetes, implemented APIs with FastAPI, and tracked experiments and model versions with ClearML while logging via ELK.
Pythonsince 2023
SQL
PyTorchsince 2023
Transformerssince 2023
PEFT
Accelerate
vLLM
GPTQ
AWQ
LangChain
RAG
CatBoost
Scikit-learnsince 2023
ONNXsince 2023
Docker
FastAPI
Kubernetes
PostgreSQL
ClearMLsince 2023
Novosibirsk State University
Bachelor's Degree •
Solid-state chemistry, physical methods of analysis