Selected engineering work across language, data and computer vision.
InnotechLLM · Agents · Data
An AI analyst, built around the agent.
Refactored a data-analyst platform into a LangGraph loop that plans, uses tools and reviews its answers. Connected analysis, forecasting and visualization in one streaming experience.
LangGraph
FastAPI
DuckDB
Plotly
P50 answer latency · platform refactor
50s→3s
11 tools. One agent loop. Analysis, retrieval, memory and review.
InnotechRetrieval & evaluation
10,000+
documents in production RAG pipelines
Making knowledge retrievable.
Built document-vectorization and retrieval pipelines, including TableRAG-inspired retrieval for structured data. Added an evaluation service to measure answer quality.
LlamaIndex
Kafka
AWS S3
DeepEval
TatITneftComputer vision & OCR
0.47 → 0.31
character error rate after OCR fine-tuning
From document to usable data.
Trained YOLOv8 for passport-field detection and fine-tuned EasyOCR on prepared datasets. Took the models into a deployed FastAPI service with a Gradio interface.
YOLOv8
EasyOCR
FastAPI
Docker
Built through hands-on experience.
From computer vision research to production agent infrastructure.
Feb 2025 — Present
Innotech
ML Engineer / LLM / Agents
Production RAG, document vectorization and evaluation services with LlamaIndex, DeepEval, Kafka and AWS S3.
MCP retrieval servers and a deep-research service for iterative planning, tool use and answer synthesis.
Agent infrastructure with an AST-validated Python sandbox, FastAPI-to-React SSE streaming, JWT authentication, and tracing through Phoenix and OpenTelemetry.
Sep 2023 — Feb 2025
ITMO University
ML Engineer / AI Laboratory
Researched video-generation methods for homogeneous-object analysis and multi-object tracking.
Built computer-vision and NLP experiments with PyTorch, Transformers, PyTorch Lightning, MMDetection and OpenCV.
Mar 2024 — Sep 2024
TatITneft
ML Engineer
Trained passport-field detection and OCR models, then deployed the service with FastAPI, Gradio and Docker Compose.
Built audio-to-structured-data pipelines with whisper.cpp and llama.cpp, plus a selectable speech-to-text model pool deployed on GPU servers.
Sep 2023 — Mar 2024
TatITneft
Backend Developer
Built FastAPI services for printed and handwritten OCR, including PostgreSQL storage, authentication, uploads and asynchronous processing.
Added pytest coverage, logging and WebSocket delivery; containerized and deployed the services on Linux.
Nov 2022 — Jun 2023
ReflectME
ML Engineer
Developed a virtual try-on pipeline for a smart-mirror startup: UNet and YOLOv8 for body and clothing segmentation, and PoseNet for pose estimation.
Trained GAN and WGAN-GP models for clothing-conditioned person generation.
Curiosity at the core. Engineering all the way.
I’m Dmitry, an AI / ML Engineer with 3+ years of experience across language models, computer vision and backend systems.
My work spans the whole path: preparing data, integrating and evaluating models, building APIs, and making systems observable in production.
Further training in computer vision and NLP at DLS, MIPT; multimodal AI at AIRI Summer School; 3D robotics and energy analytics through practical workshops.