Daniyar Abykhanov

Founding Engineer

TrueMile (YC F26)

Built and productionized an agentic freight platform with three live workflows, expanding its regression suite to 1,320 tests and building evaluation systems.

• Owned the 0→1 architecture and productionization of TrueMile's core agentic AI platform, orchestrating 3 autonomous freight workflows from initial build to live operation on real fleet data: weekly planning, disruption replanning, and idle-truck spot pursuit. • Built agent orchestration and tool-driven communication workflows across email, voice, and WhatsApp, enabling autonomous load pursuit and broker negotiation; a live flow ranked 5 market loads and drafted 5 broker emails, surfacing 3 integration defects missed by unit tests.

Research Assistant

University of Delaware

Developed a climate imputation pipeline that reduced GPU inference runtime 3,500x and achieved 62% lower RMSE than kriging.

• Cut GPU inference runtime from 3,745s to 1.06s per ensemble (3,500x speedup), enabling near-real-time climate reconstruction. • Achieved 62% lower RMSE than kriging by developing a distance-aware GATv2 imputation pipeline in PyTorch Geometric across 2,664 climate grid cells.

Machine Learning Engineer

Bank CenterCredit

largest bank in Central Asia

Deployed voice AI, scoring, RAG, and summarization systems at a bank, automating 90% of outbound debt-collection calls and improving multiple operational and credit-scoring metrics.

• Built and deployed a customer-facing Voice AI Agent automating 90% of debt-collection outbound calls; led technical demos for partner banks that adopted the solution. • Lifted credit-scoring Gini 10-15% and KPI 5-10% by owning end-to-end ML delivery across production deployments. • Reduced live-call query resolution time 30-50% with a LangGraph RAG agent used by 250+ operators bank-wide. • Cut manual document-review time 25-45% by productionizing an LLM summarization agent across high-volume internal flows.

Data Science Intern

Kazakhtelecom

largest telecom operator in Central Asia

Built churn and customer-profiling analyses that reduced churn 8% and supported campaigns with 15% higher response rates.

• Reduced customer churn 8% by building a decision tree on 120K+ records to identify the top five churn drivers. • Engineered customer-profiling datasets in BigQuery SQL, enabling campaigns that increased response rates 15%.

Master of Science, Data Science

University of Delaware

Bachelor of Science, Data Science

International Information Technology University