Jul 2025 — Oct 2026
Founder
RoleCatalystIQ · Dubai, U.A.E
- Design and deliver industrial-grade AI products end to end
- Ship POCs and MVPs from problem statement to production
- Build scalable AI pipelines and keep model performance and cost honest
About
From teaching Python in Visakhapatnam to shipping agentic AI products out of Dubai. I build the whole thing — ingestion, features, models, orchestration, interface and deployment — because the hard problems in applied AI live in the seams between those layers.

MSc Data Science graduation · Middlesex University Dubai
I started as a Python trainer, which is still the most useful thing I ever did for my engineering. You cannot teach fifty students a concept you only half understand, and you learn very quickly that a clear explanation and a correct implementation are the same skill wearing different clothes.
Analytics inside Apparel Group taught me the second lesson: a model nobody acts on is a cost centre. I was producing workforce analytics and retention predictions for people whose real question was always “so what do I do on Monday?” That question has shaped everything I have built since.
An MSc in Data Science at Middlesex University Dubai gave me the statistical grounding, and since founding CatalystIQ I have been building production systems in domains where AI has to earn its place against a spreadsheet and an experienced human: ship chandling, maritime business development, Shopify merchandising, restaurant operations, quantity surveying, inbound phone calls.
I ran CatalystIQ from July 2025 to October 2026. What I want next is depth over breadth — an AI engineering team where the problems are large enough that architecture decisions matter, and where I am responsible for systems in production rather than prototypes in a notebook.
No mock data
If a number appears in a product I built, it was computed from real data. Placeholder values have a way of surviving into production and being believed.
Structure before automation
Entity resolution, verification and a canonical schema come before any agent or model. Automating on unresolved data produces confident mistakes faster.
Interpretable beats impressive
Gradient boosting with good features over a heavier model nobody can interrogate. An operator will only act on a prediction they can question.
Cost is a design constraint
Token spend, API calls and infrastructure get budgeted at design time. An AI feature that quietly burns money is a liability, not a feature.
Teaching Python, then analytics inside a large retail group, then an MSc, then building products. Each step is on the resume and the dates match it.
Jul 2025 — Oct 2026
CatalystIQ · Dubai, U.A.E
Apr 2025 — Jun 2025
Younoh Media · Dubai, U.A.E
Feb 2025 — Apr 2025
Apparel Group · Dubai, U.A.E
2023 — 2024
Middlesex University · Dubai, U.A.E
2021 — 2023
Scholars Tuition · Visakhapatnam, India
2018 — 2023
JNTU-Kakinada · Visakhapatnam, India