I build the AI systems people actually keep using.
Lead data scientist in Dubai, available now. Nine years from marketing science for global brands to the AI infrastructure other teams build on.
Beane
Why it matters
A model nobody uses is a model that did not happen.
I started in consulting, so I learned that early. Most of my career has been spent in the gap between "the notebook works" and "the business changed what it does."
- 9 years
- 4 companies
- 5 industries
~29%Search precision
More relevant results for AI agents, inside a sub-millisecond budget. No added latency.
−90%Manual analysis
Weeks of reading customer reviews replaced by a ranked list of what to fix, with evidence attached.
40→23%Customer churn
Segmentation, a 7-day propensity model and timed retargeting, on a base of five million customers.
Selected work
Three layers, one discipline.
One from each layer I have worked at: the infrastructure other teams build on, the product a customer touches, and the decision a business makes with money.
01
Nebius · AI infrastructure
Nebius: better answers, same speed
AI search relevance. The filtering layer that decides which search results are worth putting in front of an AI assistant.
~29%more precise, no added latencyRead the case →︎
02Momos · AI product
Momos: review replies in minutes
Automated customer replies. An AI that writes on-brand answers to public reviews, and reliably refuses to touch the sensitive ones.
95%+reply rate within 24 hoursRead the case →︎
03Yum! Brands · Marketing science
Yum! Brands: pricing backed by evidence
Menu pricing and promotions. The models and the simulator behind pricing decisions for Pizza Hut and KFC in three markets.
3markets: Canada, Brazil, UKRead the case →︎
Also in production Revenue forecasting platform Customer insight engine Customer retention programme
Lab · built in the open
I built Beane, a deep research harness that ranks #1 globally on DeepResearch Bench II.
And #2 on the original DeepResearch Bench. One question in, a cited report out: it reads a hundred sources and checks every quote against its page in code. Built from scratch, its research runs on small and mid-sized models.
DeepResearch Bench II#1
60.52, 17.4 points ahead of OpenAI o3 Deep Research.
DeepResearch Bench#256.14, 0.58 points behind first.
Efficiency98%of model calls on the smallest tier. No frontier model in its research.
Beane: the rankings and the harness →︎ Currency study →︎ Official leaderboard ↗︎
How I work
Three rules I do not bend.
They exist because each one has cost me something to learn. They are also the fastest way to tell whether a project is real.
01
Agree upfront on what proves the idea wrong.
Written down before any modelling starts. It is the cheapest way to kill a bad project early.
02
Never ship without an evaluation.
If there is no way to tell whether the output is good, there is no way to tell whether it got worse.
03
Give stakeholders something they can poke at.
A simulator beats a slide. People trust a number they were allowed to argue with.
Background
Marketing science, then AI product, then infrastructure.
Founders and C-level stakeholders in every role. Solo on the AI work, sole technical lead in the marketing-science years.
Oct 2025 - May 2026AI infrastructure
Nebius (acq. Tavily)
Lead Data Scientist · Abu DhabiThe layer other teams build agents on: search relevance inside a sub-millisecond budget, a versioned revenue platform at a million records a day, and twelve-month forecasting with scenario simulation.
May 2022 - Sep 2025AI product
Momos
Lead Data Scientist (AI) · Abu DhabiTook LLM products from prototype to production for multiple brand clients: a guarded RAG response agent, AI Insights, and the LLM stack migration behind both - designed and built solo, end to end.
Aug 2017 - May 2022Marketing science
Yum! Brands & Kvantum
Lead / Senior Data Scientist · DelhiPricing and promotion analytics for Pizza Hut and KFC across Canada, Brazil and the UK, reporting into the Chief Analytics Officer. Before that, four years of consulting on marketing mix modelling, attribution and retention for brands across CPG, retail and pharma.
Notable brands
The names on the other side of the work.
Some I worked with directly on pricing and marketing science. Others I reached through the AI products I built, where the brand count runs into the hundreds.
Toolkit
What I reach for.
Applied AI & ML
LLMs, RAG, agentic systems, prompt engineering, forecasting, segmentation, churn, price elasticity, marketing mix modelling, XGBoost, CatBoost, neural networks, SHAP
Data engineering
Python, SQL, PySpark, Snowflake, BigQuery, Databricks, Airflow, AWS Lambda, Azure Data Factory, Docker, FastAPI, CDC, SCD2, Tableau, Power BI, Omni
Search & measurement
BM25, hybrid retrieval, ranking and reranking, Elasticsearch, pgvector, evaluation harnesses, A/B testing, convex optimisation, simulation, Kalman filters
Education
Post Graduate Diploma, Applied Statistics
B.Tech, Computer Science Engineering
Certifications
Nebius AI Leader Certification
AI Agents in LangGraph
Building & Evaluating Data Agents
Contact
Hiring for data, ML, or AI systems?
Immediately available, based in Dubai, and I reply fast. Happy to walk through any case in detail: the decisions, the trade-offs, and what I would do differently today.
