I am a Data scientist working at the Stepstone group with specialization in AI in NLP
Extensive research and analysis on job listings and their feature extraction to help recruiters generate listings that satisfy job requirements.
Continued pretraining, fine-tuning and evaluation of LLMs (mT5, Llama2 (german continued pretrained)) for automatic (job-)listing creation in different downstream tasks (data-to-text generation, RAG).
Optimization of training process of LLMs using PEFT methods (LoRa, QLoRa) to save time, cost and prevent over-fitting of models.
Development of RLHF-pipelines to create RLHF datasets to improve LLMs performance using text-edit events (Kafka) and monthly manual evaluation of generated texts.
Improvement of job title search by developing semantic search pipelines with Vector DBs on AWS (Milvus) to support recruiters write better job titles.
Training of semantic models.
Deployment on Sagemaker with custom docker images.
Deployment of Milvus DB on EC2 cluster.
Optimization of performance of LLMs with distributed training and inference of LLMs (DP, DDP, MP) using Pytorch, Deepspeed.
Testing and benchmarking of quantization techniques (BnB, GPTQ, AWQ) in deployment and inference of LLMs.
Building of custom inference docker images for LLMs on AWS ECR for deployment in production (AWS ECS).
AWS cloud-infrastructure design and testing using Terraform with Bamboo CI/CD tool for the purposes of RLHF (s3, AWS lambda, Dynamodb and cloud watch) and sagemaker endpoint deployment with Huggingface’s models and docker images (Text-generation-inference and Text-embedding-inference on ECR).
LLMs development including data preparation, analysis, training and testing / NLP products development / advanced analytics of Big data/ cloud deployment of machine learning models and infrastructure for cloud services
Simulation of dielectric relaxation, ferroelectric switching and imprint in PVDF copolymers (c++).
Solving non-linear differential equations to predict the behaviour of PVDF copolymers under electric fields (c++).
Experimental data visualization (Matlab).
Design of ETL pipelines to clean and populate databases (RDS) with millions of users textual data on AWS-Cloud (Pyspark, Pandas, AWS-lambda, AWS-fargate).
Implementation of merging pipelines to merge users data from different platforms (Linked-in, Stack-overflow,etc) into one single database (Dynamo-Db).
Development of an English and German model for NER (named entity recognition) to extract software skills from users information (bio,CV,etc).
Development of a matching model using BERT’s embeddings to appropriately match candidates to jobs.
Exploratory Data Analysis on hundreds of Gigabytes of scraped user’s data (numerical and textual).
Analysis of github-users’ repositories to track history and development of software skills and provide visual representation of those to recruiters.
Research in the field of Quantum Repeaters.
Design of the first Quantum Repeater with trapped ions based on single atom-photon entanglement.
Optical setup design for single ions imaging, laser beam addressing and florescence collection.
Optical and optomechanical modelling (Zeemax).
Design of digital and analog control systems to stabilize laser frequency, intensity and phase.
Experimental sequence design (Verilog(FPGA), Labview, Matlab).
Data Analysis and statistical modelling (Python (pandas), Matlab).
FEM simulations of RF Paul traps (Comsol Multiphysics).
Monte-Carlo simulation of long-distance quantum communication (C++).
The AI-first Recruitment Operating System that unifies and automates the entire hiring lifecycle for modern enterprises worldwide.
Dr Job FZ LLC. 2026 © All Rights Reserved
