Irina Lebedeva, PhD
About
I'm an AI researcher and engineer with 7+ years across the field, spanning both research and production. My work is backed by a PhD in Computer Science earned in China and 100+ citations on Google Scholar in generative modeling and computer vision. My background also includes startup experience building AI apps, and earlier years as a software engineer with team leadership — experience that still shapes how I build and ship.
Today I fine-tune large language and vision-language models and deploy inference services that stay fast and stable at scale. I'm at my best closing the gap between research and product — turning a promising result into a system people can actually rely on.
Experience
- Fine-tune LLMs for automotive requirements engineering using Qwen and LoRA.
- Drove a 30% relative gain in accuracy via synthetic data, hyperparameter tuning, and pipeline work.
- Deploy scalable LLM inference pipelines with vLLM, LoRAX, and Docker.
- Built and deployed a lightweight multi-label image-tagging model trained on synthetic VLM/LLM labels.
- Fine-tuned VLMs with Megatron-DeepSpeed (tensor + pipeline parallelism, ZeRO-3) for large-scale tagging and captioning.
- Designed a low-shot, identity-preserving image generator (LCM-SDXL + IP-Adapter + ControlNet).
- Developed text-to-video pipelines using WAN-based models and LoRA.
- Shipped a face age-estimation service (fine-tuned AuraFace, served via FastAPI).
- Improved conditional face synthesis with greater diversity using Stable Diffusion and LoRA.
- Built an AI onboarding assistant with LangChain, GPT-4, RAG, ChromaDB, and FastAPI.
- Fine-tuned LLaMA for scientific-paper summarization and Q&A.
- Published explainable CV research (hyperbolic deep network, ViT-based interpretable embeddings).
- Built VAE- and GAN-based models for image editing, makeup transfer, and masked-face inpainting.
- Shipped real-time emotion- and face-recognition systems with PyTorch, OpenCV, and Docker.
- Delivered forecasting models with XGBoost and scikit-learn; advanced facial-attractiveness research and released a labeled dataset.
- Owned the full software development lifecycle, designed SQL databases, and mentored engineering teams.
Education
PhD, Computer Science — East China University of Science and Technology
MSc, Computer Science — Shanghai Jiao Tong University
BEng, Computer Science & Engineering — National Research Nuclear University MEPhI
Selected Publications
Technical Stack
AI / ML: Transformers, Diffusion, GANs, LLMs, VLMs, RAG, LoRA
Frameworks: PyTorch, Hugging Face, DeepSpeed, Megatron, vLLM, ONNX, TensorRT
Engineering: Python, FastAPI, Triton, Docker, Git, Linux, CI/CD, AWS, SQL
Experimentation: Weights & Biases, NumPy, Pandas, scikit-learn, OpenCV
Languages
English (fluent) · Russian (native) · Chinese (professional)
Contact
Open to collaborations in generative AI, multimodal learning, and applied research.
dr.irina.lebedeva@gmail.com