Luka Mladenovic
Verified Expert in Engineering
Machine Learning Engineer and Developer
Belgrade, Serbia
Toptal member since September 1, 2026
Luka is an AI engineer with 5 years of experience delivering end-to-end AI/ML/LLM systems across high-scale SaaS and finance. He has a proven track record of taking high-stakes projects from 0 to 1 across ML and NLP, and of adapting LLMs for domain-specific use, adding significant business value. Luka is a pragmatic problem-solver with excellent communication skills, adept at collaborating with stakeholders and diverse business and engineering teams to deliver impactful AI solutions.
Portfolio
Experience
- Artificial Intelligence (AI) - 6 years
- Natural Language Processing (NLP) - 6 years
- Python - 6 years
- Machine Learning Operations (MLOps) - 6 years
- Large Language Models (LLMs) - 4 years
- Retrieval-augmented Generation (RAG) - 4 years
- Knowledge Graphs - 4 years
- AI Agents - 3 years
Preferred Environment
Python, Machine Learning, Large Language Models (LLMs), Artificial Intelligence (AI), Agentic AI
The most amazing...
...AI system I've built is the anti-abuse model at Chess.com—eliminating 99% of manual investigations and reducing abuse by 60% across a 260 million user platform.
Work Experience
Senior Machine Learning Engineer
Chess.com
- Designed and executed forward-looking Trust & Safety AI/ML roadmaps, integrating emerging technologies to reduce abuse and drive engagement.
- Spearheaded the company's first ML model in production, building end-to-end real-time anti-abuse ML/LLM systems at scale that eliminated 99%+ of manual investigation and reduced abuse levels by 60%.
- Built in-house data labeling with custom interfaces and LLM-powered synthetic data generation and augmentation, leading the human-in-the-loop process and managing a team of 10 reviewers.
- Trained and deployed models: a TF-IDF and Random Forest username classifier and a multilingual fine-tuned BERT chat transformer quantized to INT8, handling 100 requests per second at under 100 milliseconds p99 latency.
- Developed an LLM-powered abuse report system investigating non-gameplay reports at scale, clearing a 500,000+ backlog and cutting average processing time from 9 days to under 1 hour.
- Delivered agentic RAG-based Q&A systems and a knowledge-graph platform, unifying Notion, Slack, and GitHub sources in Neo4j with multi-step LangGraph agents, reporting to C-level stakeholders.
- Drove the MLOps initiative, evaluating industry-leading frameworks and leading adoption of Kubeflow, MLflow (model registry), KServe (serving), and a feature store to enable scalable AI/ML/LLM training, deployment, monitoring, and lifecycle handling.
- Led an R&D project as a technical project manager, a transformer-based recommendation model predicting move likelihood that outperformed original paper benchmarks.
Founding AI Engineer
Pathfinder
- Set the technical vision and architecture, adapting hedge-fund trading and analytics technologies to real-world problems beyond financial services.
- Led development of a project-deliverability platform for large enterprises, surfacing hidden risks and dependencies that helped clients detect margin erosion and quantify resource overallocation across programs and teams.
- Architected foundational ontology schemas, ingestion pipelines, knowledge-graph construction, graph-based ML, anomaly detection, forecasting, and downstream LLM-based Q&A.
- Constructed a knowledge graph over unstructured data from PDFs, Excel, Jira, Teams, and office software, unifying scattered enterprise sources.
- Built GraphRAG over Neo4j with fine-tuned GNN and LLM models, using Prize-Collecting Steiner Tree retrieval and GNN layers during LLM fine-tuning to achieve 2x accuracy over standard multi-hop retrieval baselines.
- Applied dominant graph anomaly detection to surface risks, combining PGExplainer with LLM-based reasoning for explainable risk detection.
- Delivered a QA interface and proactive risk detection with sandbox AI agents, spawning by severity classification for human-in-the-loop intervention.
- Worked across highly regulated domains with large enterprises, upholding the highest security standards.
ML Engineer
Veles Securities
- Developed an ML system for structuring placements and institutional sales and trading, enabling algorithmic selection of high-return opportunities for companies and investors.
- Collaborated with biotech hedge funds to map their due diligence process and factor weightings, translating that domain expertise into security-type and premium/discount recommendation models.
- Built a neural network classification model for categorical security-type recommendation and a regression model for premium/discount prediction.
- Applied NLP techniques to numerically represent unstructured textual deal components and point-in-time market sentiment as model features.
- Led data strategy across a multitude of proprietary, PIPE transaction, and financial datasets, handling acquisition negotiations, imputation, and modeling of both conventional and alternative sources.
ML Engineer
Hedge Fund
- Expanded an internal investment-research platform to ingest and process hedge-fund reports and macroeconomic datasets, increasing its coverage for due diligence workflows.
- Developed an intelligence knowledge-graph engine combining financial and public data to identify emerging topics, semantic patterns, and public-opinion signals.
- Built an enterprise-scale document processing platform handling 10,000+ financial documents and 100+ page files weekly, combining OCR with VLMs across PaddleOCR and Docling.
- Built a public-opinion analysis pipeline over Reddit and Twitter APIs, structuring temporal-aware representations and leveraging FinBERT to extract sentiment and trending tickers.
- Automated data ingestion and reporting workflows with AI agents, reducing manual processing across financial intelligence and investment-research pipelines.
ML Engineer
Episode 1 Ventures
- Built a proprietary data-driven investment platform whose deal-sourcing algorithms generated 35% of all deals in the fund’s latest vintage.
- Designed and deployed ML models that analyzed startups’ complete digital footprints across conventional and alternative datasets to identify systematically underpriced, high-potential founders.
- Developed pre-seed and seed quantitative founder-personality ML models combining 40+ behavioral variables to identify high-potential founders ahead of the broader market.
- Built a pitch-deck model using OCR and VLM to parse and quantify PDF decks with deterministic algorithms, converting unstructured decks into structured signals.
Experience
Physician Recruitment & Survey Targeting Platform
https://www.kolgroups.com/market-research-surveys/Education
Bachelor's Degree in Software Engineering
School of Electrical Engineering, University of Belgrade - Belgrade, Serbia
Certifications
Neo4j Certified Professional
Neo4j
Deep Learning with Tensorflow
IBM
Structuring Machine Learning Projects
DeepLearning.AI | via Coursera
Skills
Libraries/APIs
PyTorch, XGBoost, TensorFlow, SpaCy, React, Reddit API, X (formerly Twitter) API, PyTorch Geometric (PyG), PaddleOCR, Crunchbase API, LinkedIn API, vLLM
Tools
Named-entity Recognition (NER), Claude, Claude Agent SDK, GraphRAG, Terraform, Grafana, BigQuery, Visual Language Models (VLMs), ChatGPT, Claude Code, ARIMA, Docling, Open Neural Network Exchange (ONNX), Kueue, Flink
Languages
Python, Java, SQL
Frameworks
LangGraph, Agentic Frameworks, DSPy, Locust, Spark, Flask, Optuna, Metaflow
Paradigms
Synthetic Data Generation, DevOps, Model Context Protocol (MCP), Anomaly Detection, ETL
Platforms
Docker, Jupyter Notebook, Amazon Web Services (AWS), Kubernetes, Langfuse, LangSmith, Kubeflow, Weights & Biases, Google Cloud Platform (GCP), KServe, Azure, Confluent Kafka
Storage
Data Pipelines, Neo4j, Redis, PostgreSQL
Other
Machine Learning Operations (MLOps), Natural Language Processing (NLP), LangChain, Knowledge Graphs, AI Agents, Large Language Models (LLMs), Retrieval-augmented Generation (RAG), OpenAI, Vector Databases, Machine Learning, Data Science, Software Engineering, Problem Solving, Document Processing, Optical Character Recognition (OCR), Executive Summaries, Random Forests, Feature Engineering, Generative Artificial Intelligence (GenAI), Multimodal GenAI, LoRa, Data Gathering, Human-in-the-loop (HITL), Data Augmentation, LLM Fine-tuning, Sandbox to Production, ML Pipelines, Text Classification, Intent Classification, Classification Algorithms, Regression, Heuristics, Large Language Model Operations (LLMOps), AI Model Training, Model Evaluation, Applied Machine Learning, Model Tuning, Decision Intelligence, Debugging, Evaluation, Artificial Intelligence (AI), Agentic AI, Model Development, Communication, Applied AI, AI Engineering, Multi-agent Systems, Observability, RAG Architecture, RAG Systems, AI Modeling, Orchestration, API Design, PDF, Early-stage Startups, PDF Scraping, Agentic RAG Systems, GCP, Prometheus, MLflow, Modal, Unsloth, DeepEval, Graph Neural Networks (GNNs), Enterprise AI, Ontologies, Transformer Models, Data Engineering, Recommendation Systems, Data Strategy, Explainable Artificial Intelligence (XAI), Data Mining, Real-time Data Pipelines, Sentiment Analysis, Topic Modeling, Due Diligence, Open-source LLMs, Data Analytics, Financial Market Data, Neural Networks, LLM Integration, AI Project Management, Data Scientist, Large-scale Production Deployments, Deep Learning, Benchmarking, Product Engineering, Distributed Systems, Anthropic, Leadership, Feasibility Studies, Quantization, Performance Engineering, Employee Upskilling, Training Workshops, Microsoft Azure, Kafka, NVIDIA Triton, ONNX Runtime, Pricing Models, Argo CD, Hedge Funds, Light LLMs, Fine-tuning, CRM, Market Segmentation, Fuzzy Logic, Email Delivery, Deepeval
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