Aleksandr Shuravin

AI/ML • Middle • Izhevsk, Russia • UTC+04:00 • Remote/Hybrid • Work permits: Russia • Available in 2 weeks • from $52k per year
ML Engineer
25+ years exp
10+ years ML exp
C
C#
PHP
C++
SQL
Python
Active 17 days ago
+7 (905) 8755504
Invite to interview
Message
Download CVCV
Overview
Technical skills
Timeline
Roles

Overview

A pragmatic generalist developer at a solid junior-to-middle level who builds end-to-end prototype applications combining vision, speech and LLM integrations. The strongest proven skill is practical computer vision and application wiring, evidenced by the VisionService and the image-processing Engine and steps (megabax/ar-ai-assistant/src/vision/detectors.py and megabax/CVContainer/Libraries/Core.py). There is little or no evidence of custom model research, production-grade serving, experiment tracking or measured optimization in public code.

Technical skills

C
C#• Junior • 25y+
PHP• Junior • 25y+
C++• Junior • 25y+
SQL• Middle • 25y+
Python• Senior • 10y+
C++
PyTorch C++
Python
FastAPI
Databases
MS SQL
MySQL
MinIO• 10y+
PostgreSQL• 10y+
Qdrant• 5y+
AI/ML
Computer Vision
LLM
Transformers
NumPy
TF-Keras
OpenAI SDK
Speech Recognition
OpenCV• 10y+
PyTorch• 10y+
YOLO• 10y+
Sentence-Transformers• 7y+
Chain-of-Thought
ChatGPT
DeepSeek
Llama
Mistral
Ollama
RAG
DevOps
Git
Docker• 10y+
Rest API• 10y+
Grafana
Prometheus

Timeline

Software Engineer (ML/AI Research) • Middle
Freelance / Project work • Contractor
Jan 2025 to Present 1 Year 7 Months Izhevsk Remote/Hybrid
Built a RAG-based system for financial markets that combines news and technical analysis, performs vector retrieval, and uses LLMs to generate trading signals. Worked with Qdrant and embedding models and used DeepSeek, ChatGPT, and local Llama/Mistral models via Ollama. Added prompt design for structured outputs and implemented quality and hallucination evaluation. Also designed ML pipeline infrastructure for ETL orchestration and monitoring using Airflow, Prometheus, Grafana, Docker, Python, and PostgreSQL.
Python
PostgreSQL
Qdrant
Sentence-Transformers
ChatGPT
DeepSeek
Llama
Mistral
Ollama
Chain-of-Thought
RAG
Prometheus
Grafana
Docker
Backend R&D Developer (Video Analytics) • Middle
Programming Store • Full-Time
Nov 2016 to Mar 2026 9 Years 4 Months Izhevsk In office
Developed backend R&D for an intelligent video analytics system using deep learning and computer vision. Refactored legacy code to speed up delivery of new detectors and built a microservice for video storage, accounting, and archiving. Implemented parking detection and metal defect detection using YOLO-style approaches, and created an OCR pipeline for extracting text from technical drawings. Integrated duplicate-item search with vector retrieval (Qdrant/embeddings) and connected it to 1C via REST API; used PostgreSQL, MinIO, and Docker.
Pythonsince 2016
PyTorch
OpenCV
YOLO
PostgreSQLsince 2016
Dockersince 2016
MinIO
Qdrantsince 2016
Sentence-Transformerssince 2016
Rest API
Programmer / Full-stack Developer • Middle
Freelance / Project work • Contractor
Jan 2001 to Dec 2015 14 Years 11 Months Izhevsk In office
Delivered automation solutions for an аптечная network, retail operations, and an industrial enterprise. Implemented 1C-based functionality for warehousing, retail, and benefits workflows and added data exchange with many cash registers. Built analytical reports to improve cost structure and optimized the 1C cost calculation algorithm to reduce execution time. Integrated 1C with external services via HTTP using C#, C++, PHP, and SQL.
C#
C++
PHP
SQL
Middle AI/ML Engineer Confidence: Medium Generalist
A pragmatic generalist developer at a solid junior-to-middle level who builds end-to-end prototype applications combining vision, speech and LLM integrations. The strongest proven skill is practical computer vision and application wiring, evidenced by the VisionService and the image-processing Engine and steps (megabax/ar-ai-assistant/src/vision/detectors.py and megabax/CVContainer/Libraries/Core.py). There is little or no evidence of custom model research, production-grade serving, experiment tracking or measured optimization in public code.
Model Architecture & Training
2/10
How well models are designed and trained
Application-level use of models and basic training example; no custom architectures or training pipelines.
Evidence
megabax/ar-ai-assistant/src/ai/assistant.py: AIAssistant.answer and _call_ollama_http showing LLM inference integration and fallback megabax/CVContainer/OtherExamples/CNNMnistDemo.py: simple Keras model.fit training loop for MNIST
Data Pipeline & Feature Engineering
3/10
How data is prepared for models
Modular image-processing pipeline and utilities for frame processing and step composition.
Evidence
megabax/CVContainer/Libraries/Core.py: Engine.process orchestration of ImageProcessingStep chain megabax/CVContainer/Libraries/ImageProcessingSteps.py: MedianBlurProcessingStep, ContourProcessingStep implementations megabax/CVContainer/Exec/run2.py: assembly of processing steps and custom filter (my_filter)
Experimentation & Evaluation
2/10
How results are measured and tested
Minimal experimentation and testing artifacts; basic smoke tests and a toy train/evaluate example but no experiment tracking or reproducible pipelines.
Evidence
megabax/ar-ai-assistant/tests/test_smoke.py: SmokeTests contains unit-style checks for core components megabax/CVContainer/OtherExamples/CNNMnistDemo.py: train/evaluate flow that prints test accuracy
MLOps & Deployment
2/10
How models are shipped to production
Some operational scripts and runtime checks for models and services; no production serving, monitoring or CI/CD evidence.
Evidence
megabax/ar-ai-assistant/tools/download_vosk_model.py: scripted download, extraction and installation checks for offline speech model megabax/ar-ai-assistant/src/ai/assistant.py: base_url resolution, probing of Ollama and HTTP fallback handling
Computational Efficiency
2/10
How efficiently computing resources are used
Some attention to runtime parameters and lightweight CPU-friendly choices but no measured optimization or GPU/quantization work.
Evidence
megabax/ar-ai-assistant/src/vision/detectors.py: model.setInputParams(size=(416, 416), scale=1/255.0) for YOLO inference tuning megabax/ar-ai-assistant/src/voice/speech.py: selection and handling of a small Vosk model and explicit sample-rate/chunk handling
Research Depth & Innovation
1/10
Depth of research and new ideas
No evidence of novel algorithms, paper implementations or research-grade experimentation.
Evidence
megabax/CVContainer/OtherExamples/CNNMnistDemo.py: standard MNIST tutorial example megabax/CVContainer/Exec/run2.py: heuristic my_filter for contour approximation (handcrafted logic)
Expertise
Computer Vision & Image Analysis• Middle
Audio & Speech Processing• Middle
Conversational AI & Chatbots• Middle
Technologies
Python• Senior • 10y+
C++• Junior • 25y+
MySQL
PostgreSQL• 10y+
Rest API• 10y+
ChatGPT
OpenCV• 10y+
Qdrant• 5y+
DeepSeek
Sentence-Transformers• 7y+
YOLO• 10y+
MinIO• 10y+
FastAPI
Prometheus
Computer Vision
Chain-of-Thought
Speech Recognition
Ollama
OpenAI SDK
MS SQL
Llama
Mistral
Transformers
NumPy
Git
PyTorch• 10y+
Docker• 10y+
Grafana
LLM
RAG
TF-Keras
PyTorch C++
Computer Vision• mentioned only
Vision• mentioned only
Recommendations
  • Develop desktop or prototype AR assistants that combine camera-based detection with local/offline speech and lightweight LLM inference.
  • Build proof-of-concept computer vision pipelines and tooling for rapid prototyping (image processing step library, detector integrations).
  • Implement speech-enabled UI features or offline speech-recognition utilities for CPU-constrained environments.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories:
ar-ai-assistant AgentsDemo CVContainer