Work

Experiment design, causal measurement, uplift estimation, switchback tests, and practical methods for separating correlation from intervention effects.

LLM-powered workflows and agentic systems for real product surfaces: structured planning, tool use, evaluation harnesses, and keeping human judgment in the loop.

Applied machine learning for marketplace decisions: pricing, churn prevention, lifecycle modeling, customer and supply behavior, and operational trade-offs.

Applied ML for pricing, customer and supply behavior, and operational trade-offs in marketplace decisions.

Practical methods for separating correlation from intervention effects: uplift estimation, regression-based analysis, and causal thinking for product decisions.

A developer workflow library for AI-assisted coding. Structured planning, implementation, and review skills — built so developers stay in control while leveraging Claude.

• GitHub repo
• Read the launch post

A personal project exploring how AI tutors can help kids learn — built on the OpenClaw platform — specifically, whether an LLM-powered system can teach a child effectively while keeping human judgment in the loop. The project investigates how to structure conversations, handle misconceptions, and build genuine understanding rather than surface-level pattern matching. Currently in early-stage experimentation and not yet shareable publicly.

Agent-based simulation of tech firms choosing between human, AI-augmented, and automated production. Interactive Streamlit dashboard embedded from Hugging Face Spaces.

• GitHub repo

Experimentation platform tech lead. Built QUIK dynamic pricing, captain churn model, and dispatch optimization across Careem's Middle East super-app.

Led a DS team across subscriber lifecycle, device pricing, and operational ML. Shipped +10% subscriber inflow via retail-network footprint modeling, +3–7% gross margin on device pricing, and +15% effective ad-spend via marketing-mix modeling.

Network science and social-media research at UPF Barcelona — large-scale social-network data collection, filter-bubble measurement, and the graph-theoretic foundations underlying my current causal work.

Quantitative portfolio management with retail-credit risk components — credit scoring, default modeling, exposure management. Earliest risk-adjacent ML work.

Lecturer / instructor at MegaFon internal academy, Mail.RU Technopark, and SF Education. Modules on introductory ML, linear models, base ML methods, and experiment design (A/B testing, uplift modeling).