Данило

DANIEL BESAHA
AI Engineer | RAG, LLM Systems & Machine Learning
GMT+3 | [open contact info](look above in the "contact info" section) | [open contact info](look above in the "contact info" section)
WhatsApp | Telegram | LinkedIn
ABOUT
AI Engineer with 3.4 years of experience building production AI, machine learning, and retrieval systems. My recent work has focused on
RAG and LLM applications: document processing, embeddings, vector search, hybrid retrieval, reranking, metadata-aware search,
grounded answer generation, evaluation, and FastAPI integration. I work across the AI application lifecycle from data preparation and
indexing to retrieval quality, latency optimization, deployment support, and production monitoring.
EXPERIENCE
TRIAGO Engineering | Brașov, Romania
AI / Machine Learning Engineer | May 2023 - Present
Internal Document Intelligence / RAG
Internal AI assistant for contracts, procurement, transportation, accounting, and other operational documents, focused on reliable search
and source-grounded answers.
• Built and improved a production RAG pipeline with embeddings, vector databases, BM25, metadata filtering, hybrid retrieval, and
cross-encoder reranking.
• Used LangChain and LlamaIndex for document processing, retrieval orchestration, LLM integration, and source-grounded answer
generation.
• Improved retrieval Recall@5 from 72% to 85% by comparing dense, BM25, and hybrid retrieval, tuning top-k and filtering, and adding
reranking on the same internal evaluation set.
• Reduced unsupported LLM answers by about 40% with relevance thresholds, evidence checks, citation requirements, and an explicit
no-answer policy when context was insufficient.
• Reduced wrong-document and outdated-version retrieval through section-aware chunking, canonical entities, metadata filters, and
version-aware document handling.
• Worked with Qdrant and Pinecone for vector search and retrieval experiments, including similarity thresholds, metadata filtering,
chunking strategy, and reranking depth.
• Integrated AI/ML components with FastAPI-based services and worked in Dockerized environments for local development, testing,
debugging, and production handoff.
Lead Generation ML System
Sales automation system for identifying and prioritizing B2B leads for an engineering company supplying industrial equipment to food,
beverage, and chemical producers.
• Improved lead detection recall from 61% to 73% at about 70% precision by moving from individual-post decisions to profile-level
evaluation using text, metadata, activity, and visual information.
• Reduced expensive profile enrichment by about 44% with a lightweight first-stage filtering pipeline while preserving around 92% of
relevant leads.
• Built and evaluated classification and ranking pipelines using scikit-learn, CatBoost, TF-IDF, embeddings, feature engineering,
clustering, and PCA.
• Used MLflow and production monitoring tools for experiment tracking, error analysis, and model/pipeline quality monitoring.
SKILLS
AI Engineering & LLM: LangChain, LlamaIndex, RAG, LLM APIs, prompt engineering, grounded generation, document processing,
chunking, context management
Retrieval & Vector Search: Qdrant, Pinecone, embeddings, vector search, BM25, hybrid search, metadata filtering, cross-encoder
reranking, RAG evaluation
Backend & Integration: FastAPI, REST APIs, Python, SQL, PostgreSQL, Pydantic
Machine Learning: scikit-learn, CatBoost, XGBoost, LightGBM, PyTorch, classification, ranking, clustering, PCA, feature engineering,
cross-validation
MLOps & Production: Docker, MLflow, DVC, Kubernetes, Kubeflow, Grafana, Kafka, Git, Pytest, AWS
EDUCATION
Priazovskiy State Technical University, Dnipro
Bachelor's degree, Automation and Computer Technology | 2021 - 2025
Degree officially recognized in Romania.
LANGUAGES
English - B2 • Ukrainian - Native • Russian - Native