Mashkhura Eshmatova

I’m a machine learning engineer with a strong academic foundation in applied machine learning for finance and statistics for decision-making. I completed an MSc at University College London (UCL) and earned a BSc in Computer Science, with coursework across machine learning, probability & statistics, algorithms, linear algebra, and databases.

At Alif Uzbekistan, I build and run production ML systems end to end, from experimentation to deployment support. I architected a production chatbot using a BERT intent classifier plus RAG, supporting 3,000+ daily chats under <4s latency and automating 80% of customer support queries.

I’ve also delivered measurable impact through A/B testing and evaluation-driven iteration. My RAG system improved task success by +17% and increased automation by +11% in production experiments; I built multilingual NER (XLM-RoBERTa + CRF) and improved an OCR-pipeline NER F1 from 70% to 95% using spaCy-Transformer. I additionally designed an AI shopping agent with tool use and multi-step planning, and fine-tuned Qwen with SFT + LoRA/QLoRA using an evaluation suite for helpfulness, format adherence, and grounding.

Today, I’m leading end-to-end development of a voice assistant (ASR, NLU, TTS) for debt collection, targeting a 40% reduction in collection costs while handling 130,000 calls/month without agent involvement. I’m driven by data-driven decision-making, rigorous experimentation, and mentoring junior engineers while supporting technical hiring.