Illia Dykanskyi

Led five engineers to deliver a modular KYC platform covering document OCR/extraction, liveness, face matching, deepfake and tampering detection; shipped a production-ready MVP in one month and later replaced selected vendor flows with modules from this platform.
Built a streaming replica of 2.5+ TB from production MySQL to BigQuery, which offloaded analytical workloads and cut complex analytical queries from days or weeks to seconds; unified fragmented analytics jobs in BigQuery and created centralized database documentation, greatly reducing onboarding time.
Implemented antifraud measures to protect the platform from bot abuse, with no false-positive bans confirmed through heuristics and support data.

Remote

PythonPyTorchTensorFlowHugging FaceASRTTSChatskyRAGLangChainLangGraphLlamaIndexMCPChromaDBOpenCVCLIPRayDockerKubernetesJenkinsELKLoki

Participated in development of DeepPavlov's Dream - an open-source multimodal dialogue platform, creating new services, annotators, distributions, skills and connectors.
Performed ASR, TTS, and audio-captioning research, and built NER/LLM-based intent/entity detection with time-aligned audio processing.
Built a high-load ETL pipeline that transformed records from multiple sources into canonical entities for a recommendation core with elements of RAG, vector search, and ranking, serving both the product and an LLM chatbot with persistent memory and custom MCP tools (early adoption).
Distilled CLIP to 50% of its original size with only 2–3% accuracy loss and built multimodal agents using VLMs, streaming inputs, and recursive goal decomposition.

Remote

PythonC++C#OpenCVYOLOEfficientDetDeepSORTOpenPoseStable Diffusion 1.4PyTorchTensorFlowSageMakerOptunaMLflowffmpeg

Delivered an edge conveyor counter with <1% error at ~30 FPS on 1080p video and fine-tuned production fire, smoke, PPE, and prohibited-symbol detectors.
Automated large-volume image labeling with an OpenPose C++ pipeline and managed Stable Diffusion fine-tuning experiments across proprietary datasets.
Standardized evaluation across YOLO v3–v5, EfficientDet, and DeepSORT and open-sourced a C# annotation tool.

Open source

01UkrainianNative

02RussianNative

03EnglishProfessional working

04GermanElementary