Stanislav Liashkov

ML Engineer

Zypl.ai

R&D Lab with a focus on outlier-resistant AI applications in finance, government and education.

Adapted LLMs for Tajik with UNICEF, building training, deployment, and evaluation pipelines.

• Worked on Project SoroLLM in collaboration with UNICEF, adapting large language models to low-resource Tajik and implementing an end-to-end LLM training pipeline with continued pretraining using LoRA, instruction fine-tuning, and Direct Preference Optimization (DPO). • Built text filtering and cleaning pipelines and generated synthetic data using proprietary and open-source LLMs. • Set up, maintained, and monitored training infrastructure on AWS SageMaker and an on-premise server with 8xH200 GPUs.

AI Engineer

TWS Labs (AiXpert)

B2B AI-driven Legal Compliance System for Russian Enterprises.

Built legal RAG and evaluation systems; improved embedding-model MAP by 20.6%.

• Built a knowledge base of 200k+ Russian legislative documents and case law, comprising 3M+ chunks in Qdrant Vector DB, enriched with metadata for search and filtering. • Implemented a legal assistant workflow that retrieves legislation and case law through multiple steps with corrective reasoning, using LangChain, LangGraph, Qdrant, and LLaMA 3. • Trained a legal-document embedding model on synthetic triplets, improving MAP by 20.6% on a ranking benchmark.

ML Engineer

Relog.ai

Transport management system provider specializing in last-mile logistics.

Built a Text-to-SQL copilot that saved approximately 182.6 analyst-hours per month, plus data pipelines and a traffic model.

• Developed an in-house Text-to-SQL AI copilot for analysts using RAG, tool-calling, and LLaMA 2, reducing BI report delivery time and routine queries and saving approximately 182.6 analyst-hours per month. • Built data pipelines integrating CRM data from Bitrix24 to Power BI. • Built a baseline model for Almaty road-graph traffic density, combining time series, LISA, and supervised ML to forecast edge congestion.

Junior Data Scientist

AI Flow LTD

Consulting company delivering data-driven solutions in BI.

Contributed to churn modeling and built ETL, data-quality, and drift-detection frameworks.

• Contributed to a churn-prediction model using LightGBM, Polars, SQL, Optuna, and MLflow to support targeted marketing strategies for a client. • Implemented ETL pipelines with Python, dbt, and Apache Airflow for data-driven decision making. • Built automated data-quality and drift-detection frameworks for ML inference using Great Expectations, Pandas, and SciPy statistical tests.

Master’s degree, Data Science

University of Colorado Boulder

Bachelor equivalent, Computer Science

42 Bangkok