09/11/2026 updated
RM
Premium member
100 % availableSenior AI & ML Engineer | GenAI, LangGraph, Python Backend | 10+ Years
Sofia, Bulgaria
Only remote
MSc in Quantitative Financial Modelling, University of Pantheon Sorbonne, ParisAbout me
Hands-on AI Software Engineer specializing in the entire ML lifecycle. With 10+ years of enterprise experience, I focus on building scalable backend architectures, designing RAG pipelines, and deploying production-ready generative/multi-agent systems.
AI AgentsArgo CDRAGGitHub CopilotGitHub ActionsMLflowLLMOpenAIFine-TuningGCPAzure DevOpsLangChainPython (Programming Language)Machine LearningNumPy
AI & Architecture
Expertise in LLM architectures, RAG, Agentic Systems, LangChain, LangGraph, Vector Databases, TensorFlow, PyTorch, Scikit-learn, NLP, Deep Learning, and Generative AI. Design and deployment of production-grade predictive and GenAI solutions including multi-agent systems and continued pretraining.
MLOps & LLMOps Engineering
Proficiency in Python, Docker, Kubernetes, ArgoCD, Helm, MLflow, CI/CD, SQL, FastAPI/Flask, A/B Testing, Drift Detection, and Model Retraining. End-to-end management of ML and LLM lifecycles including fine-tuning, prompt engineering, and inference optimization.
Cloud Infrastructure & Deployment
Hands-on experience with Azure (Databricks), AWS (Bedrock), Google Cloud (Vertex AI, GKE), Serverless Computing, and Local/On-premise Inference using NVIDIA DGX, A100s, and vLLM. Scalable and low-latency AI solution deployment across major cloud platforms.
End-to-End Ownership
Proven ability to take complex AI requirements from PoC to production-ready scalable systems with minimal supervision, focusing on robust backend engineering and clean code.
NLP Pipelines & Transformer Models
Development of state-of-the-art NLP models for translation (Roberta2Roberta), summarization (Pegasus/BART), Sentiment Analysis, NER, POS, and Question Answering. Training of diverse transformer architectures from scratch for specialized languages.
Agentic Systems & RAG Pipelines
Design and deployment of advanced RAG pipelines and agentic services as APIs, extracting unstructured data from complex documents using both small and large open-source models. Orchestration of complex multi-agent LLM systems for autonomous data retrieval and multi-step reasoning.
Quantitative Analysis & Financial Modeling
Research and development of intraday quantitative strategies for equity trading at a hedge fund. Background in quantitative financial modelling with an MSc from the University of Pantheon Sorbonne.
System Integration & API Design
Robust API design ensuring scalability and low-latency performance, seamlessly integrating agentic pipelines and predictive models into legacy microservice landscapes. Management of databases and cloud infrastructure using Docker, Kubernetes, and custom APIs.
LLMOps & Synthetic Data Generation
Utilization of tools such as Magpie and Meta Llama family to generate high-quality synthetic data. Application of model merging techniques to enhance domain-specific capabilities including math proficiency.
Advanced Inference & On-Premise Hardware
Extensive experience with cloud (Runpod) and on-premise hardware (A100s, H100s), utilizing vLLM and text-generation-inference for optimized deployment of large language models.
Experience
Professional experience0 - 5 years
Freelancer since2026
Languages
EnglishFluent
Project history
Researched, developed and deployed production-grade Smart FBT (Frequently Bought Together) models integrating predictive intelligence into enterprise vending ecosystems. Designed and orchestrated complex multi-agent LLM systems for autonomous data retrieval and automated enterprise Q&A services. Acted as primary AI technical authority collaborating with cross-functional teams. Managed end-to-end LLM lifecycle including local fine-tuning, prompt engineering, and inference optimization. Drove AI strategy bridging product requirements and technical execution. Ensured scalability and low-latency performance through robust API design integrating agentic pipelines into legacy microservice landscapes. Designed and analyzed A/B experiments for the Smart FBT recommendation engine.
Trained diverse transformer architectures from scratch for the Bulgarian language, available on HuggingFace. Utilized Magpie and Meta Llama family to generate high-quality synthetic data and applied model merging techniques. Extensive experience with cloud and on-premise hardware (A100s, H100s) using vLLM and text-generation-inference. Developed state-of-the-art NLP models for translation, summarization, Sentiment Analysis, NER, POS, and Question Answering.
Key creator of Calcatron, a highly accurate predictive pricing engine deployed globally, listed in the US patent under creator. Designed and deployed advanced RAG pipelines and agentic services as APIs for extracting unstructured data from complex financial and operational documents. Conducted daily training, evaluation, and advanced prompt engineering on open and closed-source models. Deployed ML models in production with robust monitoring, logging, and data drift detection. Managed databases and cloud infrastructure (Google Cloud) using Docker, Kubernetes (ArgoCD/Helm), and custom APIs. Implemented closed-source LLMs for automated call-center operations.