Laurentiu Piciu
Verified Expert in Engineering
Machine Learning and Software Developer
Bucharest, Romania
Toptal member since November 14, 2022
Laurentiu is an AI and data engineer who builds AI agents that turn messy documents into structured data, and the data platforms they run on. He led the technology behind an AI platform for private-credit analysts, an AI agent that took over most of a pharma intelligence company's client service, and a zero-downtime BigQuery-to-ClickHouse migration that cut data costs by more than half. He believes persistence and an analytical mind matter most in the AI era.
Portfolio
Experience
- Artificial Intelligence (AI) - 9 years
- Microservices - 9 years
- Python - 9 years
- Machine Learning - 9 years
- Natural Language Processing (NLP) - 9 years
- Chatbots - 5 years
- ChatGPT - 4 years
- DeepSeek - 1 year
Work Experience
Lead Data Engineer
High-Traffic Consumer Mobile App
- Led the migration of the data platform from BigQuery to ClickHouse on AWS, cutting data infrastructure costs from $45k to $20k a month.
- Built real-time event ingestion with dbt, Airflow, Spark, Iceberg and S3, managed as code in Terraform and Ansible, replacing pay-per-query billing with capacity the team controls.
- Put every part of the data pipeline under control: alerts, dashboards, and memory and cost tracked for every dbt model and every dashboard.
- Built a platform AI agents can work on without limits: fast queries in ClickHouse, no per-query bill for each question an agent asks, and documentation written for agents across 2,000+ tables.
- Cut the daily memory load of the ClickHouse pipelines by 71% in four weeks on the same hardware: out-of-memory query failures went from 18 a day to zero, and dashboards with a broken tile from 1 in 15 to 1 in 500.
Chief AI and Data Scientist
US Private Credit AI Startup
- Led the technology behind a new AI-driven platform that changes how investment analysts work with their data: instead of reading credit documents and retyping figures into spreadsheets, they query structured, verified data.
- Designed and shipped the AI agents that extract and reconcile metrics from credit agreements, offering memoranda, agent notices and financial statements, with every value traceable to its source page.
- Owned the whole technical stack end to end: the analyst-facing frontend, the backend and APIs, and the cloud infrastructure on Azure, defined in Terraform.
- Designed a tenant-per-client architecture in which each client's documents and data run in an isolated cloud environment, to meet the security requirements of institutional investors.
- Built the analyst workbench that shows extracted metrics, the source PDF and the calculation behind each value side by side, so a number is verified in one click.
- Opened the extracted data to AI assistants through an MCP server, so analysts ask questions about their portfolio in plain language.
AI Engineer
LATAM Pharmaceutical Market Intelligence Platform
- Took a pharmaceutical market intelligence company from dashboards only to an AI agent its clients can question directly, because dashboards never answered every client question.
- Designed and built an AI agent that answers market questions in natural language over the company's pharmaceutical data, used by 1,000+ industry professionals.
- Offloaded most of the client service workload to the agent, leaving the data and client service teams free for deeper analysis and client work.
- Made the conversational agent better release by release: analysed its conversations, defined the recurring problems in its answers, and fixed them one by one.
AI Developer
Circles
- Built an AI travel concierge agent on OpenAI models that understands travelers' requests and calls flight, hotel and other travel APIs to return real answers.
- Designed the FastAPI microservice architecture behind the agent, from request handling to the travel API integrations.
- Created a configuration system that lets the internal team change the agent's behavior and parameters without code changes.
Senior AI Developer
Anguleris Technologies
- Trained the LLM to generate reliable 3-part specification documents for construction products.
- Designed and developed the entire Azure architecture for document storage, pre-processing, analytics, and training.
- Developed the API that provides the service for the internal team.
AI Developer
LB.AI Inc.
- Developed an ecosystem that connects to a database and allows users to ask natural language queries to get immediate responses from the database.
- Developed a Telegram bot that integrates with the conversational ecosystem and allows the interaction of the user with the database.
- Integrated OpenAI's LLMs (prompt design, API development) for translating natural language queries into SQL code.
Staff ML Engineer | Data Scientist
Lummetry.AI
- Collaborated as a part of the founding team that built a startup from zero until the leading security services providers in Romania acquired it.
- Defined ML/DL architectures and implementation plans for all proofs-of-concept (PoC), custom services, and products we developed.
- Designed, built, trained, and deployed into production neural models for predictive analytics, natural language processing (NLP), and computer vision.
- Implemented ML/DL pipelines for a distributed AI-based near-real-time safety and security system. Automated the surveillance system in bank branches.
- Led the predictive analytics area for online retail recommendations and automatic category management.
- Implemented predictive analytics systems for diminishing the out-of-stock and overstock burden by predicting sales and discovering duplicate payments.
Undergraduate Teaching Assistant
University Politehnica of Bucharest
- Acted as the teaching assistant for the graduate course "Data Science with Deep Learning."
- Provided teaching assistance for the undergraduate laboratory "Object Oriented Programming."
- Prepared laboratory materials, support, exercises, and home assignments. Helped the students to solve problems and offered constructive feedback.
Project History
Legal Documents Processing
• Exposes the functionalities via REST APIs to be integrated into the client's ecosystem
• Drastically reduced human costs that manually performed the tasks
Distributed AI-based Near-real-time Safety and Security System
• The purpose of this product is to gather data from multiple streams (CCTV cameras, video files, sensors). I analyzed all the data in near-real-time using deep neural models to apply some algorithms for raising alerts when needed (e.g., crowded zones, people fallen or in distress, obstructed camera, etc.) or just performing some distributed jobs (e.g., blurring a video file).
• Allows deep neural models to be continuously improved by training them with data collected in production.
• Communicates with the IoT platform that integrates all the alerts via MQTT.
LENS+
• Evaluated demand and sales forecasting based on complex non-linear patterns.
• Identified customer needs changes and constructed data-driven automatic insights.
• Integrated into retail solutions as processes that train models on the client's data and outputs results which are used to take some actions.
PaidAnalytix
• Developed a neural model to identify duplicate payments.
• With the use of siamese neural networks, the model could cut the quantity of manual labor by 90% while maintaining a 95% recall rate.
• Analyzed all payments in an Excel file and output the duplicate pairs.
MultiGatedUnit
https://ieeexplore.ieee.org/document/9324853• The architecture aimed to reduce the ample experimental time and computational resources due to the search process of the architectural design.
Education
Bachelor's Degree in Computer Science
University Politehnica of Bucharest - Bucharest, Romania
Certifications
Artificial Intelligence Professional Program
Stanford University
Skills
Libraries/APIs
TensorFlow, Keras, Pandas, NumPy, Matplotlib, OpenAI API, React, PyTorch, SpaCy, Scikit-learn, Natural Language Toolkit (NLTK), REST APIs, Hugging Face Transformers, OpenCV, Node.js
Tools
GitHub, PyCharm, ChatGPT, Claude, Apache Airflow, Apache Iceberg, Terraform, Ansible, Grafana, Claude Code, Claude Agent SDK, Pytest, Jupyter, Docker Compose, TensorBoard, Named-entity Recognition (NER), RabbitMQ, Azure Machine Learning, Azure Kubernetes Service (AKS), Google AI Platform, Celery, DeepSeek
Languages
Python 3, SQL, Bash, Python, Unicorn, TypeScript, Scala, Java, Regex, C++, HTML, CSS
Frameworks
Apache Spark, LangGraph, Agentic Frameworks, Flask, Django, Streamlit
Paradigms
ETL, Testing, Microservices, Object-oriented Programming (OOP), Siamese Neural Networks, Automation, Business Intelligence (BI), Model Context Protocol (MCP), Test-driven Development (TDD), Continuous Deployment, Continuous Integration (CI)
Platforms
Linux, MacOS, Azure, Google Cloud Platform (GCP), Azure AI Studio, Docker, Windows, NVIDIA CUDA, Jupyter Notebook, AWS Lambda, Amazon Web Services (AWS)
Storage
MongoDB, PostgreSQL, ClickHouse, Amazon S3 (AWS S3), Data Pipelines, NoSQL, Databases, Azure Cloud Services, MySQL
Other
Machine Learning, Software Engineering, Algorithms, Deep Learning, Artificial Intelligence (AI), Data Science, Natural Language Processing (NLP), Predictive Analytics, Computer Vision, Optimization, Linear Algebra, Directed Acrylic Graphs (DAG), Transfer Learning, Data Analytics, Research, Servers, Data Visualization, Performance Tuning, Performance Testing, Text Classification, Recommendation Systems, ML Pipelines, Object Tracking, Deep Neural Networks (DNNs), Convolutional Neural Networks (CNNs), Image Annotation, AI Design, Image Processing, Distributed Systems, Processing & Threading, Data Analysis, Generative Pre-trained Transformers (GPT), Chatbots, OpenAI GPT-4 API, OpenAI GPT-3 API, Predictive Modeling, Automated Data Flows, Scripting, Back-end Development, Large Language Models (LLMs), Language Models, OpenAI, API Integration, APIs, Generative Pre-trained Transformer 3 (GPT-3), English, Prompt Engineering, LangChain, AI Agents, Finance, Retrieval-augmented Generation (RAG), Amazon Bedrock AgentCore, Model Tuning, Serverless, Vector Search, AI Chatbots, Vector Databases, Google BigQuery, Data Build Tool (dbt), Data Migration, Generative Artificial Intelligence (GenAI), AI Workflow, Conversational AI, Engines, Knowledge Bases, Agentic AI, AI Engineering, Agentic RAG Systems, LLM Agents, Large Language Model Operations (LLMOps), RAG Architecture, Observability, Agentic AI Systems, LLM Evaluation, AI/ML Workloads, Semantic Search, CI/CD Pipelines, Telemetry, Solution Architecture, AI Architecture, Operating Systems, Statistics, Big Data, Calculus, Web Scraping, Data Engineering, Object Detection, Object Classification, Time Series, Machine Learning Operations (MLOps), Leadership, Graphics Processing Unit (GPU), BERT, OOP Designs, Data, Regular Expressions, Pinecone, Weaviate, Gemini, Capital Markets, Trading, Open-source LLMs, Meta Llama, Pgvector, Google ADK, Reinforcement Learning, Deep Reinforcement Learning, MLflow, GitHub Actions, Decentralized Systems, FastAPI, Multiprocessing, Slackbot, Interactive JavaScript, HTML Integration, Data Scraping, Decision Intelligence, AI Testing, Text-to-text Transfer Transformer (T5)
How to Work with Toptal
Toptal matches you directly with global industry experts from our network in hours—not weeks or months.
Share your needs
Choose your talent
Start your risk-free talent trial
Top talent is in high demand.
Start hiring