AI/ML engineer in Abu Dhabi. I build and deploy detection models, usually where the interesting failures are rare and unlabelled. On one project I trained a one-class model on healthy sensor data only and reached 99% failure recall across 220K readings. I placed 1st in the UAE and 4th internationally out of 123 teams at the Dragon Oil AI Hackathon.
I work in Python with MLflow, Optuna, Docker, and FastAPI, and I care about reporting metrics a reviewer can actually trust. When a model scores suspiciously well, I want to know why before it ships.
Open to
AI/ML Engineer roles and internships in the UAE.
Spoken languages
Languages
ML & Deep Learning
MLOps, Cloud & Deployment
| Domain | Proficiency | Details |
|---|---|---|
| Anomaly Detection & Predictive Maintenance | Strong | One-class / unsupervised modelling on industrial sensor data; honest model selection over inflated scores |
| Time-Series Forecasting | Strong | Demand and call-volume forecasting; time-based splits to avoid leakage |
| Classical ML | Strong | scikit-learn, XGBoost, LightGBM, Random Forest; evaluation metrics and tuning with Optuna |
| Computer Vision | Proficient | YOLOv8 detection, tracking, and counting in real-time pipelines |
| MLOps & Reproducibility | Proficient | MLflow experiment tracking, reproducible pipelines, containerized deployment |
| Robotics / Autonomous Systems | Proficient | ROS, ESP32 and MQ-135 sensor integration, real-time telemetry |
SensorGuard — Predictive-Maintenance Anomaly Detection
Flags failing pumps from 52 sensor channels, trained on healthy readings only. Compared eight models in a reproducible pipeline, then chose Isolation Forest for realistic deployment because the near-perfect supervised scores were detecting recovery rather than predicting failure. Shipped as a Streamlit dashboard with a live gauge, per-sensor interpretability, and batch scoring.
| Stack | Python, scikit-learn, Streamlit, Plotly |
| Scale | 220K time-series rows, 52 sensors |
| Performance | 99% failure recall, scored in under two seconds |
| Tooling | Reproducible benchmark across 8 models, batch CSV scoring |
| Impact | Catches equipment failure early, with metrics a reviewer can trust |
| Repository | github.com/yashicode/sensorguard |
UAE EMS Demand Forecasting
Forecasts Dubai's next-month ambulance call volume to within roughly 9% (MAE about 1,500 on a 12,000-call base) using official DCAS open data from 2017 to 2024. A simple Linear Regression beat ensemble models on the 83-month series, and a time-based split exposed how a random split's inflated 97% R-squared misleads on time-series data.
| Stack | Python, scikit-learn, Streamlit, Docker |
| Scale | 83-month series, official open data (2017-2024) |
| Performance | MAE ~1,500 on a 12,000-call base (~9%) |
| Tooling | Containerized dashboard with live forecasts and a what-if tool |
| Impact | Demonstrates honest time-series evaluation over a misleading random-split score |
| Repository | github.com/yashicode/ems_response |
Metro Exit Counter — Computer Vision
Real-time pipeline that detects, tracks, and counts people exiting a defined zone in metro footage. Per-ID tracking avoids double counting, and the exit-zone geometry is resolution independent. A live overlay shows count, FPS, and crowd density, tuned for throughput on modest hardware.
| Stack | Python, YOLOv8n, OpenCV |
| Scale | Real-time video, per-ID multi-object tracking |
| Performance | Live counting with FPS and density overlay on modest hardware |
| Tooling | Resolution-independent zone geometry |
| Impact | Footfall counting for transit-style environments |
| Repository | github.com/yashicode/metro-exit-counter |
AI-Driven CO2 Emission Detection Drone — Senior Capstone
ROS-based autonomous drone integrating ESP32 microcontrollers with MQ-135 gas sensors for real-time CO2 monitoring. Built an end-to-end sensor-to-decision pipeline with a real-time React/Node.js dashboard. Selected for the DEWA Cleantech 2025 Hackathon.
| Stack | ROS, ESP32, MQ-135, React, Node.js |
| Scale | Live flight with continuous environmental sampling |
| Performance | 95% data reliability, emission accuracy of ±50 ppm, sub-second dashboard latency |
| Tooling | Sensor-to-cloud analytics with live alerts |
| Impact | Autonomous environmental monitoring without a manual pilot |
| Repository | github.com/yashicode/drone-co2-monitoring-app |
Dragon Oil AI Hackathon — Production Flow Optimization
Production-flow optimization model for oil and gas, built for the Dragon Oil AI Hackathon (GOTECH 2025). Placed 1st in the UAE and 4th internationally out of 123 teams. Logged 40+ runs in MLflow with full metrics, hyperparameters, and artifacts, and automated tuning with Optuna for a 20% gain over manual tuning.
| Stack | Random Forest, XGBoost, LightGBM, MLflow, Optuna |
| Scale | 123 teams internationally |
| Performance | 1st in UAE, 4th worldwide; +20% over manual tuning |
| Tooling | 40+ tracked MLflow runs, Power BI monitoring on the MLflow backend |
| Impact | Top-ranked solution under competition constraints |
| Repository | github.com/yashicode/Random-Forest-Model-for-Oil-Gas-and-Water-Production-Forecasting |
Machine Learning Intern · Cognifyz Technologies · Remote
Jan 2026 - Mar 2026
- Built demand-forecasting models for food-service operations with Python, scikit-learn, and XGBoost, improving forecasting accuracy by 35% for scheduling and resource planning.
- Automated ETL data pipelines with Python and SQL, cutting manual processing and raising operational efficiency by 50%.
- Partnered with operations and business teams to build Power BI dashboards used for day-to-day decisions.
Python scikit-learn XGBoost SQL Power BI
Software Engineer Intern · Digital Value Stream · Dubai, UAE
Jun 2024 - Oct 2024
- Worked in a 6-person agile team to ship React and Next.js dashboards for a live-event platform.
- Documented requirements for core modules, improving navigation and user engagement by 25%.
React Next.js Agile
| Recognition | Details |
|---|---|
| 🏆 Dragon Oil AI Hackathon (GOTECH 2025) | 1st in the UAE, 4th internationally among 123 global teams |
| 🥈 Nova Dubai Entrepreneurship Pitch Summit | 2nd Place, University of Dubai Pitch Competition |
| 🥈 Microsoft x Amity x Seneca Polytechnic Design Jam | 2nd Place internationally |
| 🚀 RTA UAE Innovates 2024 | Finalist: IoT platform for real-time vehicle data, predictive maintenance, and alerts |
| 🎯 MDX Business Case Challenge | Top 5 |
Microsoft
-181717?style=for-the-badge&logo=github&logoColor=white&labelColor=0D1117){kind=icon}
AWS
-232F3E?style=for-the-badge&logo=amazonwebservices&logoColor=white&labelColor=0D1117){kind=icon}
DataCamp
Anthropic
Other
-5E5E5E?style=for-the-badge&labelColor=0D1117){kind=icon}
building: anomaly-detection and forecasting models that deploy
exploring: MLOps, computer vision, autonomous systems
learning: Microsoft AI/ML certifications
open_to: AI/ML Engineer roles & internships (UAE)
Building machine learning systems that hold up when someone checks the math.