Location: Tbilisi, Georgia, with plans to move to Serbia/Hungary
Remote: Yes
Willing to relocate: No
Technologies: LLMs, GPT, Llama, transformers, Instruction Tuning, Preference Tuning, Alignment, DPO, LoRA, RAG, PyTorch, Hugging Face, DeepSpeed, Qdrant, vLLM, Triton Inference Server, TensorRT, Prompt Engineering, Python, FastAPI, Docker, GCP, MLFlow, Redis, PostgreSQL, Git, CI/CD, mypy, flake8, rye, pytest, Semantic Search, LLM-as-a-Judge, Rust, Kotlin, Ktor
CV: https://docs.google.com/document/d/1C-KXW0JUgzfkVe3nOmF9wI0b...
Email: bobazooba@gmail.com
My name is Boris and I have been actively involved in dialogue systems (both in the pre-LLM era and with LLM models) and everything related to language processing for more than 7 years.
At my last job, I developed an internal framework for generating complex synthetic data. Goal: very precise fine-tuning of LLMs and encoders for complex, multi-stage tasks, including RAG, scenario following, goal progression, content promotion, and so on. This approach helped solve 6 LLM-based projects in the company.
I also developed and implemented a zero-shot intent classifier that works with an open list of classes and determines where to go next in the scenario based on dialogue. Essentially, it's an instructional classifier where intents are described in free form.
I have extensive experience in evaluating the quality of generative models and am familiar with most of the pitfalls in this process. In total, my portfolio includes more than 20 different LLM-based projects.
Besides ML, I have strong engineering skills and experience in writing fairly complex backend code in Python for AI agents. I researched and developed a scenario engine based on LLM. I also have open source experience, with a library for simplifying LLM fine-tuning that has nearly 400 stars: https://github.com/BobaZooba/xllm
Earlier, I taught Deep Learning in NLP for 4 years at Russia's best university.
Currently, I am actively looking for a new job. I have much more to share. I would be happy to discuss and come to a fruitful collaboration in the LLM field.