Looks like you have JavaScript disabled. For the full Twine experience, you will need to re-enable it.

I’m a Senior AI Engineer who builds production-grade LLM-powered systems, specializing in RAG (retrieval-augmented generation) and agentic AI architectures. I’ve led end-to-end development across data pipelines, model training, retrieval and orchestration, evaluation, and observability—designing solutions that are reliable, low-latency, and cost-efficient for real-world business needs. From legal summarization engines to enterprise conversational AI and fraud/risk modeling, I translate complex, ambiguous domain problems into robust, measurable products. I work across AWS and GCP, using modern ML/MLOps and LLMOps tooling to deploy scalable inference, validate output quality, and continuously improve performance through automated benchmarks and monitoring.…I’m a Senior AI Engineer who builds production-grade LLM-powered systems, specializing in RAG (retrieval-augmented generation) and agentic AI architectures. I’ve led end-to-end development across data pipelines, model training, retrieval and orchestration, evaluation, and observability—designing solutions that are reliable, low-latency, and cost-efficient for real-world business needs. From legal summarization engines to enterprise conversational AI and fraud/risk modeling, I translate complex, ambiguous domain problems into robust, measurable products. I work across AWS and GCP, using modern ML/MLOps and LLMOps tooling to deploy scalable inference, validate output quality, and continuously improve performance through automated benchmarks and monitoring.

Mihai Dragomir

AI Engineer, Data Scientist, Software Engineer, +3





I’m a Senior AI Engineer who builds production-grade LLM-powered systems, specializing in RAG (retrieval-augmented generation) and agentic AI architectures. I’ve led end-to-end development across data pipelines, model training, retrieval and orchestration, evaluation, and observability—designing solutions that are reliable, low-latency, and cost-efficient for real-world business needs. From legal summarization engines to enterprise conversational AI and fraud/risk modeling, I translate complex, ambiguous domain problems into robust, measurable products. I work across AWS and GCP, using modern ML/MLOps and LLMOps tooling to deploy scalable inference, validate output quality, and continuously improve performance through automated benchmarks and monitoring.…I’m a Senior AI Engineer who builds production-grade LLM-powered systems, specializing in RAG (retrieval-augmented generation) and agentic AI architectures. I’ve led end-to-end development across data pipelines, model training, retrieval and orchestration, evaluation, and observability—designing solutions that are reliable, low-latency, and cost-efficient for real-world business needs. From legal summarization engines to enterprise conversational AI and fraud/risk modeling, I translate complex, ambiguous domain problems into robust, measurable products. I work across AWS and GCP, using modern ML/MLOps and LLMOps tooling to deploy scalable inference, validate output quality, and continuously improve performance through automated benchmarks and monitoring.

Available to hire

I’m a Senior AI Engineer who builds production-grade LLM-powered systems, specializing in RAG (retrieval-augmented generation) and agentic AI architectures. I’ve led end-to-end development across data pipelines, model training, retrieval and orchestration, evaluation, and observability—designing solutions that are reliable, low-latency, and cost-efficient for real-world business needs.

From legal summarization engines to enterprise conversational AI and fraud/risk modeling, I translate complex, ambiguous domain problems into robust, measurable products. I work across AWS and GCP, using modern ML/MLOps and LLMOps tooling to deploy scalable inference, validate output quality, and continuously improve performance through automated benchmarks and monitoring.

Skills

Experience Level

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Beginner

Language

Work Experience

Senior AI Engineer | Tech Team Lead at LexisNexis (Lexis+ AI)

February 1, 2024 - July 31, 2026

Led development of a production-grade Legal Summarization Engine processing long-form case law (100+ pages) into structured headnotes and holdings, reducing attorney review time by 60%+. Fine-tuned LLaMA 3 70B and LLaMA 2 13B with QLoRA to generate citation-grounded summaries, improving factual consistency by ~45%. Built large-scale SFT pipelines using Apache Spark + Delta Lake to curate and preprocess 10M+ legal documents. Designed hierarchical summarization for documents exceeding 200K tokens. Architected multi-agent workflows with LangGraph (Retriever, Summarizer, Critic) and implemented Model Context Protocol (MCP) for standardized context exchange across agents and retrieval. Integrated GPT-4o and Claude for synthetic data generation, distillation, and LLM-as-judge evaluation. Implemented hybrid RAG with Pinecone + Elasticsearch and a citation validation layer to reach >92% citation accuracy, with automated evaluation and performance optimizations using PyTorch FSDP and vLLM/Tenso

Senior AI Engineer | Tech Team Lead at LexisNexis

February 1, 2024 - July 31, 2026

Led development of a production-grade Legal Summarization Engine on Lexis+ AI, processing long-form case law into structured headnotes and holdings and reducing attorney review time by 60%+. Fine-tuned LLaMA 3 70B and LLaMA 2 13B using QLoRA to generate citation-grounded legal summaries with improved factual consistency. Built large-scale SFT data pipelines using Apache Spark + Delta Lake to curate and preprocess 10M+ legal documents. Designed hierarchical summarization for documents exceeding 200K tokens and implemented multi-agent workflows with LangGraph (Retriever, Summarizer, Critic), reducing post-editing cycles by ~40%. Implemented RAG with hybrid retrieval (Pinecone + Elasticsearch), built citation validation for jurisdiction-specific references (>92% accuracy), and established automated evaluation pipelines for faithfulness and reasoning consistency. Optimized training with PyTorch FSDP and deployment with vLLM and TensorRT-LLM to achieve 2–3x throughput gains and sub-second

Senior AI Engineer at Parloa

December 1, 2021 - January 31, 2024

Led development of enterprise conversational AI systems across voice and chat channels, improving intent recognition accuracy by 30%+ and scaling automated support for high-volume contact center operations. Designed end-to-end real-time pipelines (ASR → NLU → Dialogue Manager → TTS) to reduce average handling time by 20–25% and improve CSAT/containment. Fine-tuned BERT/RoBERTa models for intent classification and NER, improving precision/recall by ~25% across multilingual interactions. Built retrieval-based response systems using Elasticsearch + dense embeddings, and implemented multi-turn dialogue state management and memory to reduce user repetition. Integrated GPT-3.5/early GPT-4 for dynamic responses and guardrails to ensure safe, consistent, brand-aligned outputs, reducing escalation to human agents by ~35%. Optimized low-latency inference with FastAPI and deployed scalable AI services on AWS with CI/CD pipelines.

Senior Data Scientist at Revolut

September 1, 2018 - October 31, 2021

Built and deployed machine learning models for fraud detection and transaction anomaly detection, reducing fraudulent transactions by 30%+ across card payment flows. Developed real-time risk scoring for payment authorization with improved fraud precision while maintaining low false-positive rates. Designed feature engineering pipelines over large-scale financial datasets (billions of transactions), improving performance by 20%+. Implemented gradient boosting models (XGBoost/LightGBM) for credit risk and user behavior prediction, improving approval accuracy and reducing losses. Built real-time streaming analytics pipelines (Kafka + Spark Streaming) for fraud monitoring with sub-second decision latency. Developed experimentation and A/B testing frameworks, customer segmentation using clustering, and ensemble modeling/feature store architecture that increased detection recall by 25%+. Deployed models as scalable Python microservices via REST APIs.

Software Engineer at Fortech

October 1, 2010 - May 31, 2016

Developed and maintained enterprise-grade backend systems and web applications for European clients, improving reliability and reducing production incidents by ~30%. Built scalable RESTful APIs and microservice-style architectures using the Java/Spring ecosystem for high-traffic enterprise workflows. Designed and optimized relational database schemas and SQL queries, reducing query latency by 40%+. Implemented backend services for data processing and workflow automation, improving operational efficiency. Contributed to distributed system components focused on fault tolerance, logging, and observability, and integrated third-party APIs for secure data exchange. Participated in Agile development lifecycle and supported CI/CD adoption for continuous delivery.