Hi There,I'm Daniyar
I am all about |
I build AI that ships and scales — not portfolio demos. Multi-agent LLM systems and production RAG pipelines, built and deployed. Google AI Hackathon Winner
M.S. Data Science @ University of Delaware, graduating May 2026.
Featured Achievement
Google Build With AI Hackathon 2026
Daniyar Abykhanov — ML/AI Engineer
Team TreeRoute @ NYU Tandon
Presenting our architecture to the audience
Google Build With AI Hackathon 2026
All participants @ NYU Tandon
About me
Who I am
ML/AI Engineer specializing in multi-agent systems with planning, tool use, and RAG — shipped to production, measurable business impact.
Pursuing an M.S. in Data Science at the University of Delaware (GPA 3.7, graduating May 2026). Previously a Machine Learning Engineer in FinTech — owned end-to-end ML product delivery from credit scoring to LLM-powered internal tools.
Stack: Python, TypeScript, LangGraph, LangChain, PyTorch, FastAPI, React/Next.js.
Highlights
- 🏆 Google Build With AI Hackathon 2026 Winner @ NYU Tandon
- 🎓 M.S. Data Science — University of Delaware, GPA 3.7
- 💼 2+ years production ML at FinTech
- 🔬 Research Assistant — climate ML with PyTorch Geometric
Tech Stack
ML / AI
Languages & Data
Frontend
Deployment & Infra
Experience
Work History
TreeRoute
Technical Co-Founder
New York / Remote
- Won Google Build With AI Hackathon 2026 at NYU Tandon and continued development of the winning route-planning product into a production-ready AI application.
- Built a Gemini 2.5 Flash agent with tool-calling across 4 real-time APIs (Maps, Routes, Pollen, Weather) and integrated NYC tree census data for grounded recommendations.
- Led end-to-end AI product development, owning agent architecture, backend infrastructure, and production integration.
University of Delaware
Research Assistant
Newark, Delaware
- Developed and optimized a distance-aware GATv2 imputation pipeline in PyTorch Geometric for CMIP6 climate data, achieving 62% lower RMSE vs kriging (0.144K vs 0.378K) across 2,664 grid cells.
- Ran ablation experiments across 9 GATv2 architectures, narrowing the gap to GraphEM to 1% and identifying land-ocean features and multi-head attention as the most impactful design choices.
- Optimized GPU inference to reduce runtime from 3,745s to 1.06s per ensemble, enabling near-real-time climate reconstruction.
Bank CenterCredit
Machine Learning Engineer
Almaty, Kazakhstan
- Lifted 10–15% Gini and 5–10% KPI by owning end-to-end ML product delivery — scoping, feature engineering, training, validation, and production deployment of credit scoring systems.
- Cut 40–60% ML iteration time by building an automated pipeline (Python, SQL, LightGBM, SHAP, Airflow) from raw data to monitored production models.
- Shipped a multi-agent LangChain system with tool-augmented retrieval that reduced call center knowledge lookup time by 30–50%; deployed as an internal product used daily by staff.
- Reduced 25–45% manual review time by productionizing an LLM document summarization agent over high-volume internal document flows.
Kazakhtelecom
Data Science Intern
Almaty, Kazakhstan
- Built a decision tree model on 120K+ customer records to identify the top 5 drivers of churn, informing retention actions that reduced customer churn by 8%.
- Built customer profiling datasets in SQL by joining production tables and engineering segmentation features for targeted marketing campaigns.
Education
Academic Background
University of Delaware
Master of Science, Data Science
Newark, Delaware
- Focus: Multi-agent systems, RAG, deep learning, statistical modeling
- Research: Spatio-temporal climate imputation with Graph Neural Networks
Projects
What I've Built
TreeRoute
WebsiteHackathon Winner — Google Build With AI 2026
Allergy-aware walking route planner for New York City. Ranks routes by pollen exposure using NYC street tree census data, real-time weather/wind, and Gemini 2.5 Flash for natural-language explanations. Supports voice input and audio route summaries.
Knowledge Copilot
RAG system for PDF Q&A
Full-stack document intelligence platform. Upload PDFs, ask questions, get streamed answers with source citations and page references. Hybrid retrieval (vector search + BM25), analytics dashboard, and feedback loop for answer quality tracking.
AI Investment Advisor
WebsiteML + LLM financial planning system
Personalized investment strategy engine. Classifies risk profiles, detects market regimes via KMeans clustering, runs 10,000 Monte Carlo simulations for goal-probability estimates, and offers a LangGraph-powered AI copilot for follow-up questions.
Contact
Get in touch
Feel free to reach out — I'll get back to you quickly.
Direct contact
© 2026 Daniyar Abykhanov
Built with Next.js + Tailwind CSS