Ievgen – LLM, Python, SQL, experts in Lemon.io

Ievgen

From Uruguay (UTC-3)flag

Machine Learning Engineer|Strong senior

Ievgen – LLM, Python, SQL

Senior ML / Speech AI Engineer with 15+ years in speech technology and 20+ years in software engineering. Ievgen builds voice and LLM products that hold up in real-world conditions: noise-robust speech recognition for voice agents, speech LLMs for spoken language understanding, recognition of atypical speech, and pronunciation scoring for language learning. He takes models through the full cycle, from research and benchmarks to production inference. He has published at Interspeech and SLaTE and currently runs his own LLM-based call analytics platform.

Main highlights:

  • Built a call analytics platform on his own that checks 100% of a contact center's calls against its quality matrix, which human-only QA can't cover
  • Owned the full voice pipeline (STT, VAD, keyword spotting, language detection, SLU) for a drive-thru conversational AI product, from research to production
  • Built a state-of-the-art ASR system for dysarthric speech, published at Interspeech 2023
23 years of commercial experience in
Adtech
AI
Customer support
Edtech
Food and beverages
Healthcare
Hospitality
Legal tech
Machine learning
Marketing
Scientific research
Telecommunications
SaaS
Main technologies
LLM
4 years
Python
11 years
SQL
11 years
AI
16.5 years
Voice AI integration
15 years
Additional skills
RabbitMQ
Redis
MATLAB
OpenGL
AWS
GCP
Tensorflow
Pandas
NumPy
Deep Learning
Anthropic
OpenAI
Scikit-learn
Data Science
XGBoost
AI agent orchestration
RAG
Multi-agent systems architecture
AI agent development
Multi-Agent Systems
LLM integration
Direct hire
Possible
Ready to get matched with vetted developers fast?
Let’s get started today!

Experience Highlights

Founder and developer
Jul 2024 - Ongoing2 years 3 months
Project Overview

A SaaS platform for contact centers that uses LLMs to analyze calls. Manual QA usually checks only a small sample of calls, so teams miss compliance risks and can't see how each agent is really doing. Ievgen designed and built the whole platform himself. It uses agentic workflows on AWS Bedrock, OpenAI and Gemini to score every call against the client's own quality matrix (COPC), and also analyzes interviews, focus groups and open-ended survey answers. Now teams track 100% of calls, see each agent's performance, and catch compliance risks early without hiring more QA staff.

Project gallery:
Portfolio example for Uspeech by Eugene, Founder and developer
Responsibilities:
  • Built agentic call analytics workflows (COPC) that track 100% of calls and monitor each agent's performance
  • Designed and built the full platform solo: Python/Django backend, PostgreSQL, React frontend, deployed on AWS and Azure
  • Integrated multiple LLM providers (AWS Bedrock, OpenAI, Google GenAI) with LangGraph and Pydantic for structured analysis
  • Built LLM analysis of in-depth interviews, focus groups and open-ended survey answers
  • Integrated Paddle for billing and set up Docker-based CI/CD
  • Set up agentic client outreach and sales workflows with Claude, Apollo, Sales Navigator and Google Ads
Project Tech stack:
AWS
LLM
Bedrock
Microsoft Azure
GCP
Gemini API
Anthropic
LangGraph
Django
PostgreSQL
Pydantic
RAG
AI agent orchestration
Multi-agent systems architecture
Docker
CI
CD
React
Senior Research Engineer
Jan 2023 - Ongoing3 years 9 months
Project Overview

A conversational AI company that automates order-taking at drive-thru restaurants. Off-the-shelf speech-to-text models don't work well in noisy drive-thru conditions, which hurts order accuracy. Ievgen owned the whole voice pipeline: custom real-time ASR, voice activity detection, keyword spotting, language detection and spoken language understanding. He took it from research to production and kept it running there. He also fine-tuned on-prem multimodal and text LLMs for SLU and set up their inference with vLLM.

Responsibilities:
  • Built a custom real-time ASR pipeline for noisy drive-thru speech, covering data preprocessing, training, benchmarking and inference (SpeechBrain, NeMo, Transformers, Kaldi, C++)
  • Fine-tuned on-prem multimodal and text LLMs for spoken language understanding (Transformers, Unsloth) and deployed their inference with vLLM
  • Led research on adapting state-of-the-art STT models to the drive-thru domain
  • Built custom VAD, language detection and keyword spotting models with their inference pipelines
  • Built a parametrized dialogue generator for synthetic data and a training pipeline for NLU
  • Improved the STT model continuously by analyzing large volumes of production stats
  • Set up an Azure CycleCloud (Slurm) multi-GPU training cluster for the whole team, with Docker, Kubernetes and GitHub Actions
Project Tech stack:
C++
Python
LLM
PyTorch
Kaldi
Unsloth
vLLM
Docker
Kubernetes
MLflow
Microsoft Azure
Hugging Face
LLM evaluation
RAG
LLM benchmarks
CI
CD
Senior ML engineer
Nov 2019 - Jan 20233 years 2 months
Project Overview

A speech recognition app for people with non-standard speech: users with speech disabilities, aging adults and accented speakers. Standard voice tech can't understand these users, so voice assistants and smart homes don't work for them. Ievgen built the whole voice AI stack: data preparation, custom model architecture, training, inference pipeline and integration. This included a state-of-the-art ASR for continuous dysarthric speech and on-device command recognition for users with severe speech disorders. The work was published at Interspeech 2023.

Responsibilities:
  • Built a state-of-the-art ASR system for continuous dysarthric English speech, based on WavLM and a custom WFST-based decoder (PyTorch, K2, Kaldi, Icefall, SpeechBrain)
  • Designed a custom language-independent speech command architecture for users with severe speech disorders (smart home domain), with on-device C++ inference
  • Built a platform for training personalized models
  • Augmented training data with voice conversion and TTS
  • Co-authored "A personalised speech communication application for dysarthric speakers" (Interspeech 2023)
Project Tech stack:
Python
PyTorch
Kaldi
C++
Hugging Face
AWS
Microsoft Azure
Julia
Docker
CI
CD
Senior ML Engineer
Jul 2015 - Nov 20194 years 3 months
Project Overview

An ML consulting company that delivers NLP, computer vision and time series projects for clients in adtech, legal tech and entertainment. Ievgen led a team through the full ML cycle, from data collection to inference pipelines.

Responsibilities:
  • Built a computer vision model that creates object-relationship graphs from images
  • Built fact extraction from Swedish legal documents
  • Built models that predict the commercial success of entertainment events
  • Trained a custom German-English translation model
  • Built a platform for AI-powered editing of ad creatives
Project Tech stack:
PyTorch
Tensorflow
Scikit-learn
NumPy
Pandas
SQL
Python
XGBoost
OpenCV
Software Engineer
May 2003 - May 20074 years
Project Overview

Multipurpose ultrasound imaging systems for medical diagnostics. Ievgen built the core of the application software: a high-FPS pipeline for processing and rendering ultrasound data in real time, communication with the scanner hardware, and the full business logic.

Responsibilities:
  • Built a framework for real-time ultrasound processing and rendering in B, M, spectral and color Doppler modes
  • Built the UI with a custom widgets framework and the communication protocol with the control panel device
  • Built communication protocols with the scanner hardware and the scanning state logic
  • Built semi-automatic measurements, cine loop playback and import/export, and elastography visualization (C++, DirectX)
Project Tech stack:
C++
DirectX

Education

2005
Applied Mathematics
Master of Science in Applied Mathematics

Languages

Spanish
Advanced
Ukrainian
Advanced
English
Advanced

Hire Ievgen or someone with similar qualifications in days
All developers are ready for interview and are are just waiting for your requestdream dev illustration
Copyright © 2026 lemon.io. All rights reserved.