Alexander is available for hire
Hire AlexanderAlexander Chistyakov
Verified Expert in Engineering
Software Engineer and Developer
Bozeman, MT, United States
Toptal member since December 30, 2025
Bio
Alexander is a senior full-stack ML engineer with 5+ years of experience. He has worked in quantitative finance, developing ML-based trading and arbitrage strategies, as well as quantitative sports betting. He has designed, deployed, and built robust data pipelines for GBDTs, regression models, and transformers. With extensive experience in PyTorch, LightGBM, XGBoost, and numerous other models, Alexander is skilled in deploying infrastructure and creating pipelines from scratch on bare Linux.
Portfolio
Formation Partners
Databricks, Bayesian Inference & Modeling, Data Engineering, Web Scraping...
VoiceVerse Labs Ltd
Machine Learning, Machine Learning Operations (MLOps), AI Engineering...
Toptal
Artificial Intelligence (AI), Gradient Boosting, Gradient Boosted Trees...
Experience
- Pandas - 6 years
- Torch - 5 years
- Reinforcement Learning - 5 years
- Backtesting - 5 years
- Probabilistic Graphical Models - 5 years
- XGBoost - 5 years
- Sentiment Analysis - 4 years
- LightGBM - 4 years
Preferred Environment
Linux, LightGBM, PyTorch, Amazon Web Services (AWS), Google Cloud, Databricks, XGBoost, Anthropic, Pandas, PySpark
The most amazing...
...project I've worked on involved developing ML-based trading bots based on people's descriptions of their trading strategies.
Work Experience
Lead Data Engineer
2026 - PRESENT
Formation Partners
- Worked with Databricks to design, manage and orchestrate pipelines for 100GB+ of data.
- Managed and maintained web scrapers for fuzzy and incomplete wage/spending data, along with cleaning, organizing and preparing features for a Bayesian model.
- Trained, managed and created a web UI for Splink (an AI linking library) to organize over 10GB of incomplete, scraped data.
- Optimized pulling tactics to cut data ingestion times by over 60%.
- Solved bugs in external APIs and communicated with their dev teams to improve data accuracy by over 30%.
Technologies: Databricks, Bayesian Inference & Modeling, Data Engineering, Web Scraping, Amazon Web Services (AWS)
ML Platform Engineer (Data & Evals) for a Voice AI Startup
2026 - 2026
VoiceVerse Labs Ltd
- Built and owned a multi-cloud fleet system that located and provisioned scarce A100 GPU capacity across providers and regions, shipped in 2 days under an urgent deadline. Cut company fleet costs 1.5x from previous infrastructure.
- Automated worker provisioning and teardown with Kubernetes and Terraform, releasing idle GPUs between audio model inference jobs to cut wasted compute spend.
- Rebuilt the customer-facing front end in React; shipped to production as the live site.
- Cut internal evaluation dashboard load times 13x by converting blocking operations to async processing.
- Migrated issue tracking to GitHub Issues and built a framework connecting agents to that infrastructure, accelerating DataOps workflows.
- Resolved recurring CI/CD pipeline failures and automation defects, improving build reliability across the team, saving up to 35% in build time.
Technologies: Machine Learning, Machine Learning Operations (MLOps), AI Engineering, AI Model Training, Python, LLM Integration, Data Science, Artificial Intelligence (AI), MLflow, React, JavaScript, GCP DevOps, Kubernetes
Developer
2026 - 2026
Toptal
- Built a prediction model to identify the risk of failure for air conditioner parts based on geographic data, past failure rates, and venue size/purpose.
- Took charge of presenting and pitching the idea to Carrier Inc.
- Developed a baseline integration with UI elements for efficient and comprehensive model output to non-technical users.
Technologies: Artificial Intelligence (AI), Gradient Boosting, Gradient Boosted Trees, Data Science, Data Engineering
AI Engineer
2026 - 2026
Lennar Homes LLC
- Designed a multi-agentic system (BigLen) that fetches, aggregates, and reasons based on internal company information, aggregated from various sources (financial data, sales data, phone call data, and more).
- Identified a common bug class, which at least three other contractors were solving at any given time, traced its origin to the foundation, and eliminated it through core architectural changes.
- Cut token costs by 40% by greatly reducing LLM MCP tool calls through proper tool design.
- Eliminated numeric hallucinations in LLM outputs through proper structured output parsing, sanitization, and handling via tags (made agents never touch the actual numeric data).
- Implemented proper AgentCore practices and parallelization to further reduce latency and token costs (prompt caching, pre-warming microVMs, and effective parallelization in sequential code).
- Reviewed and approved pull requests from the team.
Technologies: Python, Large Language Models (LLMs), Snowflake, Model Context Protocol (MCP), Amazon Web Services (AWS), TypeScript, Amazon Bedrock AgentCore, Agentic AI, AWS Glue, Multi-agentic Systems, AI Hallucinations Management, Multi-agent Systems, Prompt Engineering
Data Scientist
2025 - 2025
Propsbot.ai
- Built and maintained data ingestion pipelines using Dagster and Apache Airflow, integrating real-time sports data from Odds API and MySportsFeeds.
- Developed player prop prediction models for NFL and NHL using XGBoost, including model calibration for online inference.
- Implemented Gaussian Mixture Models (GMM) and Hidden Markov Models (HMM) to capture player-level latent states and seasonal dynamics, improving predictive performance.
- Designed and executed A/B testing frameworks to evaluate and deploy challenger models in production.
Technologies: XGBoost, LightGBM, Scikit-learn, Pandas, Torch, Backtesting, Bayesian Statistics, Probabilistic Graphical Models, Databricks, Apache Airflow, Dagster, Bayesian Machine Learning, Spark, Spark SQL, Artificial Intelligence (AI), Machine Learning, Node.js, RESTFul APIs, TypeScript, Keras, Multi-tenant SaaS, Data Processing, Pattern Recognition, PostgreSQL, CI/CD Pipelines, Vercel, Snowflake, AWS Glue, JavaScript, Amazon S3 (AWS S3), Next.js, Data Science, ETL, Deep Neural Networks (DNNs), Model Deployment, Machine Learning Operations (MLOps), Data Pipelines, Kubernetes, Tekton, Amazon EC2, Amazon Kinesis, Amazon EKS, Pgvector, Data Analysis, Data Engineering, MySQL, Data Cleaning, SQL, Data Labeling, Data Modeling, ETL Pipelines
Data Scientist and Software Engineer
2024 - 2025
Xtrades
- Deployed and monitored Gradient Boosted Decision Tree (GBDT) models using LightGBM for predictive signal scoring, trade filtering, and model-driven decision support.
- Designed and implemented real-time financial data pipelines for market data aggregation, validation, and feature generation used across trading and analytics systems.
- Integrated news sentiment analysis pipelines to enrich trading signals, combining NLP-based sentiment features with price and volume data to improve signal robustness.
- Developed simulation and stress-testing frameworks using variational autoencoders to generate synthetic market scenarios, enabling strategy validation under rare or unseen market conditions.
- Applied statistical validation and monitoring techniques to ensure model stability, drift detection, and performance consistency in live trading environments.
Technologies: Sentiment Analysis, Variational Autoencoders (VAEs), LightGBM, Scikit-learn, Pandas, Torch, Apache Airflow, Transformers, Backtesting, Simulations, Bayesian Statistics, Model Validation, Artificial Intelligence (AI), Large Language Models (LLMs), Architecture, Back-end, Machine Learning, Natural Language Processing (NLP), Distributed Systems, Multimodal Models, vLLM, Retrieval-augmented Generation (RAG), AI Chatbots, Vector Databases, Generative Artificial Intelligence (GenAI), Azure OpenAI Service, Keras, Full-stack, Agentic AI, Amazon SageMaker, Amazon Bedrock AgentCore, Data Processing, Data Visualization, Financial Modeling, Pattern Recognition, Predictive Analytics, PostgreSQL, CI/CD Pipelines, API Integration, Vercel, Docker, Data Science, Deep Neural Networks (DNNs), RAG Architecture, NLU, Model Deployment, Data Pipelines, Kubernetes, Pgvector, Light LLMs, Data Analysis, Data Engineering, Data Cleaning, Data Labeling, Data Modeling, ETL Pipelines
Full-stack ML Engineer
2020 - 2024
Onebrain Technologies
- Built and deployed Python-based AI models for statistical arbitrage, integrated into live trading systems.
- Deployed production-grade ML pipelines using TensorFlow, Kubernetes, and Apache Airflow, achieving a 40% reduction in signal latency.
- Fine-tuned a news sentiment analysis model using DeBERTa, integrated with a Hierarchical Hidden Markov Model (HHMM) to detect market regimes and dynamically condition trading signals on macro and sentiment states.
- Developed advanced quantitative trading strategies using ensemble learning, Bayesian models, and reinforcement learning for adaptive indexing, regime-aware weighting, portfolio rebalancing, and execution optimization.
- Implemented advanced risk management using Causal Bayesian Networks, Topological Data Analysis, and Extreme Value Theory for systemic, regime, and tail-risk modeling.
- Built high-frequency execution systems with latency optimization, predictive order management, and real-time anomaly detection.
- Designed continuous adaptation frameworks, leveraging online learning, transfer learning, and evolving model ensembles to maintain performance across changing market conditions.
Technologies: Python, XGBoost, LightGBM, Torch, TensorFlow, Scikit-learn, Pandas, Transformers, Statistical Arbitrage, Bayesian Machine Learning, Sentiment Analysis, Fine-tuning, Reinforcement Learning, Variational Autoencoders (VAEs), Diffusion Models, Topological Data Analysis, Graph Neural Networks (GNNs), Causal AI, Backtesting, LSTM, FastAPI, Artificial Intelligence (AI), Large Language Models (LLMs), Architecture, Back-end, Machine Learning, AI Voice Agents, Conversational AI, Natural Language Processing (NLP), RESTFul APIs, TypeScript, Voice, VoiceBot, Twilio API, Vapi, Distributed Systems, Multimodal Models, Large-scale Distributed Systems, vLLM, Retrieval-augmented Generation (RAG), AI Chatbots, Data Protection, Vector Databases, LangChain, OpenAI, Mistral AI, HIPAA Compliance, Generative Artificial Intelligence (GenAI), Azure, Keras, Full-stack, Amazon Web Services (AWS), Multi-tenant SaaS, Agentic AI, Amazon SageMaker, Amazon Bedrock AgentCore, Data Processing, Data Visualization, Financial Modeling, Pattern Recognition, Predictive Analytics, PostgreSQL, CI/CD Pipelines, API Integration, Vercel, Model Context Protocol (MCP), Snowflake, JavaScript, Amazon S3 (AWS S3), Docker, Django, Next.js, Data Science, ETL, Deep Neural Networks (DNNs), RAG Architecture, NLU, Model Deployment, Machine Learning Operations (MLOps), Data Pipelines, Kubernetes, Amazon EC2, Amazon EKS, Amazon Textract, Optical Character Recognition (OCR), Pgvector, Supabase, UiPath, Light LLMs, Data Analysis, Data Engineering, MySQL, Data Cleaning, SQL, Data Modeling, ETL Pipelines
Experience
Player Prop Prediction and Simulation System for NFL and NHL
https://propsbot.ai/I built a production-ready player prop prediction system for NFL and NHL games, combining machine learning, probabilistic modeling, and experimental evaluation to generate accurate, real-time player performance predictions. Precise prediction of player performance is challenging due to latent player states, seasonal dynamics, and non-stationary performance trends. Simple statistical models often fail to adapt to changing conditions or provide reliable confidence estimates.
My solution involved developing a hybrid modeling pipeline that captures both observable and latent patterns in player performance. This included XGBoost-based prediction models for individual player props, calibrated for online inference to provide real-time predictions. I also leveraged latent state modeling using Gaussian Mixture Models (GMMs) and Hidden Markov Models (HMMs) to capture player-level hidden traits and seasonal dynamics, improving predictive accuracy and robustness. Additionally, I designed and implemented A/B testing frameworks to evaluate challenger models in production, ensuring continual model improvement and reliable deployment.
My solution involved developing a hybrid modeling pipeline that captures both observable and latent patterns in player performance. This included XGBoost-based prediction models for individual player props, calibrated for online inference to provide real-time predictions. I also leveraged latent state modeling using Gaussian Mixture Models (GMMs) and Hidden Markov Models (HMMs) to capture player-level hidden traits and seasonal dynamics, improving predictive accuracy and robustness. Additionally, I designed and implemented A/B testing frameworks to evaluate challenger models in production, ensuring continual model improvement and reliable deployment.
Regime-aware News Sentiment Engine for Algorithmic Trading
I designed and implemented a regime-aware news sentiment system that combines state-of-the-art NLP with probabilistic time-series modeling to condition trading signals on market context dynamically. The system improves signal robustness by adapting sentiment interpretation across different macro and volatility regimes.
My technical approach included fine-tuning a DeBERTa-based news sentiment model on financial news and market-relevant text to generate high-fidelity sentiment scores. I implemented a Hierarchical Hidden Markov Model (HHMM) to infer latent market regimes at multiple time scales (short-term sentiment shocks versus long-term macro states). I
conditioned trading signals on the joint state of news sentiment embeddings, inferred market regime probabilities, and macro and volatility features. I also integrated regime-aware sentiment features directly into downstream alpha models and execution logic.
My technical approach included fine-tuning a DeBERTa-based news sentiment model on financial news and market-relevant text to generate high-fidelity sentiment scores. I implemented a Hierarchical Hidden Markov Model (HHMM) to infer latent market regimes at multiple time scales (short-term sentiment shocks versus long-term macro states). I
conditioned trading signals on the joint state of news sentiment embeddings, inferred market regime probabilities, and macro and volatility features. I also integrated regime-aware sentiment features directly into downstream alpha models and execution logic.
AI-driven Quantitative Trading and Risk Management System
I developed an end-to-end AI-powered trading and risk management system that combines advanced machine learning, reinforcement learning, and probabilistic modeling to generate adaptive strategies and robust risk assessments for financial markets. This project involved building a multi-layered system integrating strategy generation, portfolio optimization, execution coordination, and risk management.
Synthetic indices were created using Random Forests, Gradient Boosting, and ensemble learning to generate robust signals. Bayesian Adaptive Factor Models were used for regime-aware stock weighting, dynamically adjusting allocations to changing market conditions. Reinforcement learning was used for portfolio and execution: Model-Based RL (MBRL) for dynamic portfolio rebalancing under risk and liquidity constraints, and Multi-Agent RL (MARL) for coordinated trade execution, reducing market impact and optimizing multi-asset strategies. Regarding advanced risk management, Causal Bayesian Networks were used for systemic risk inference, and Extreme Value Theory (EVT) was leveraged for tail-risk and stress-event modeling, protecting portfolios from rare but catastrophic events.
Synthetic indices were created using Random Forests, Gradient Boosting, and ensemble learning to generate robust signals. Bayesian Adaptive Factor Models were used for regime-aware stock weighting, dynamically adjusting allocations to changing market conditions. Reinforcement learning was used for portfolio and execution: Model-Based RL (MBRL) for dynamic portfolio rebalancing under risk and liquidity constraints, and Multi-Agent RL (MARL) for coordinated trade execution, reducing market impact and optimizing multi-asset strategies. Regarding advanced risk management, Causal Bayesian Networks were used for systemic risk inference, and Extreme Value Theory (EVT) was leveraged for tail-risk and stress-event modeling, protecting portfolios from rare but catastrophic events.
YourSearch.ai
YourSearch.ai is an advanced AI-powered search engine designed to offer personalized, unbiased, and summarized search results in real-time. The platform is integrated directly into messaging applications, allowing users to interact with search technology in a more seamless, natural way. The project focuses on delivering highly relevant search results tailored to the individual’s preferences and needs, enhancing the overall search experience.
Key features:
• Personalized search: Delivers customized search results based on individual preferences, ensuring a unique experience for each user.
• Unbiased results: Prioritizes accuracy and neutrality, providing users with unbiased and relevant information.
• Summarized content: Summarizes lengthy articles, research, and data, presenting only the most pertinent details for faster consumption.
• Real-time integration: Seamlessly integrates into messaging platforms to allow users to interact with the search engine without leaving their preferred communication tools.
Key features:
• Personalized search: Delivers customized search results based on individual preferences, ensuring a unique experience for each user.
• Unbiased results: Prioritizes accuracy and neutrality, providing users with unbiased and relevant information.
• Summarized content: Summarizes lengthy articles, research, and data, presenting only the most pertinent details for faster consumption.
• Real-time integration: Seamlessly integrates into messaging platforms to allow users to interact with the search engine without leaving their preferred communication tools.
Levity – AI-powered Email Automation for Logistics
https://levity.ai/Levity is transforming logistics operations through AI-powered email automation. By streamlining processes like quoting, order entry, and tracking, Levity enhances efficiency and reduces manual workloads for companies like Shine Logistics Group and Ultraship. Its customizable workflows and seamless integrations with existing tools make it a powerful ally in freight management.
Key features:
• AI-powered workflows: Automates email classification and processing for quoting, order entry, and tracking.
• Data extraction: Captures critical information from emails and PDFs, including pickup dates, weights, and destinations.
• Seamless integration: Works with TMS platforms (e.g., Rose Rocket) and rate engines (e.g., DAT) for real-time pricing and load creation.
• Customizable responses: Generates replies within 30-60 seconds, with options for drafts or fully autonomous emails.
• Versatility: Handles various input formats but may face challenges with handwritten or cursive text.
Key features:
• AI-powered workflows: Automates email classification and processing for quoting, order entry, and tracking.
• Data extraction: Captures critical information from emails and PDFs, including pickup dates, weights, and destinations.
• Seamless integration: Works with TMS platforms (e.g., Rose Rocket) and rate engines (e.g., DAT) for real-time pricing and load creation.
• Customizable responses: Generates replies within 30-60 seconds, with options for drafts or fully autonomous emails.
• Versatility: Handles various input formats but may face challenges with handwritten or cursive text.
Education
2018 - 2020
Bachelor's Degree in Mathematics
University of Colorado Boulder - Boulder, CO, USA
Skills
Libraries/APIs
PyTorch, XGBoost, TensorFlow, Scikit-learn, Pandas, LSTM, Node.js, Twilio API, vLLM, Keras, React, PySpark
Tools
Azure OpenAI Service, Amazon SageMaker, AWS Glue, Tekton, Amazon EKS, Amazon Textract, Claude Code, Apache Airflow, Spark SQL, Vim Text Editor, Hidden Markov Model
Languages
Python, TypeScript, Snowflake, JavaScript, SQL
Frameworks
LightGBM, LangGraph, Django, Next.js, Spark
Paradigms
HIPAA Compliance, Model Context Protocol (MCP), ETL, Text Retrieval
Platforms
Databricks, Azure, Amazon Web Services (AWS), Vercel, Docker, Kubernetes, Amazon EC2, Harness, Linux
Storage
PostgreSQL, Amazon S3 (AWS S3), Data Pipelines, Elasticsearch, MySQL, Google Cloud
Other
Torch, Transformers, Statistical Arbitrage, Bayesian Machine Learning, Sentiment Analysis, Reinforcement Learning, Variational Autoencoders (VAEs), Topological Data Analysis, Backtesting, Bayesian Statistics, Probabilistic Graphical Models, Simulations, Model Validation, Time Series, A/B Testing, Random Forests, Bayesian Networks, Meta-learning, Artificial Intelligence (AI), Large Language Models (LLMs), Architecture, Back-end, Machine Learning, AI Voice Agents, Conversational AI, Natural Language Processing (NLP), RESTFul APIs, Voice, VoiceBot, RAG Pipelines, Distributed Systems, Multimodal Models, Large-scale Distributed Systems, Retrieval-augmented Generation (RAG), AI Chatbots, Data Protection, Vector Databases, LangChain, OpenAI, Mistral AI, Generative Artificial Intelligence (GenAI), Full-stack, Multi-tenant SaaS, Agentic AI, Amazon Bedrock AgentCore, Data Processing, Data Visualization, Financial Modeling, Pattern Recognition, Predictive Analytics, CI/CD Pipelines, API Integration, Server-side PDF Generation, Agentic RAG Systems, Data Science, Deep Neural Networks (DNNs), RAG Architecture, NLU, Model Deployment, Machine Learning Operations (MLOps), Amazon Kinesis, Optical Character Recognition (OCR), AI Agent Orchestration, Pgvector, Supabase, Light LLMs, Data Analysis, Data Engineering, Data Cleaning, Data Labeling, Data Modeling, ETL Pipelines, AI Hallucinations Management, Multi-agent Systems, Prompt Engineering, Mathematical Modeling, Mathematical Analysis, Diffusion Models, Graph Neural Networks (GNNs), Causal AI, FastAPI, Dagster, Vapi, UiPath, Fine-tuning, Mathematics, Reinforcement Learning from Human Feedback (RLHF), Multi-agentic Systems, Gradient Boosting, Gradient Boosted Trees, Anthropic, AI Engineering, AI Model Training, LLM Integration, MLflow, GCP DevOps, Bayesian Inference & Modeling, Web Scraping
Collaboration That Works
How to Work with Toptal
Toptal matches you directly with global industry experts from our network in hours—not weeks or months.
1
Share your needs
Discuss your requirements and refine your scope in a call with a Toptal domain expert.
2
Choose your talent
Get a short list of expertly matched talent within 24 hours to review, interview, and choose from.
3
Start your risk-free talent trial
Work with your chosen talent on a trial basis for up to two weeks. Pay only if you decide to hire them.
Top talent is in high demand.
Start hiring