Ali

Data Scientist

SastaTicket.PK • Meio período

Sep 2024 - Jan 2025 • 4 mos

Project: "Dynamic-Fly" (Real-time Pricing Engine) The Story: I worked remotely with a cross-functional squad of Marketing and Product leads to solve a major pain point: static pricing in a volatile travel market. My goal was to move away from manual adjustments and toward a model that reacts to the market in seconds. The Work: I built and deployed a dynamic pricing algorithm for hotel bookings using CatBoost and SQL. To ensure the model stayed fresh, I used Airflow to orchestrate real-time data from both internal transactions and external competitor scraping. The Result: Our team saw a 12% increase in conversion rates during the peak holiday season, as our prices finally matched live market demand.

ML Engineer

Motive • Meio período

Mar 2024 - Aug 2024 • 5 mos

Project: "Safe-Drive" (Real-time Anomaly Detection) The Problem: Detecting dangerous driving patterns (harsh braking, rapid acceleration) in real-time across 10,000+ active vehicles without high latency. The Solution: Built a lightweight Lstm-Autoencoder for unsupervised anomaly detection on streaming IoT sensor data. Optimized the model using Apache Kafka for ingestion and TensorFlow Lite for edge-case validation. The Result: Reduced false-positive alerts by 40% and increased driver safety scores across the fleet by 22%.

Computer Vision Engineer

National University of Sciences and Technology • Período integral

Jan 2023 - Aug 2023 • 7 mos

Professional Summary Computer Vision Engineer at NUST, specializing in the transition of high-level academic research into production-grade AI solutions. Expert in developing real-time perception systems using YOLOv11, PyTorch, and TensorRT for defense, healthcare, and industrial automation. 1. Project: "Aegis-Vision" (Autonomous Surveillance) Tech: Multi-spectral (RGB/Thermal) fusion using YOLOv10 optimized for NVIDIA Jetson. Impact: Achieved 99.2% accuracy in real-time nighttime threat detection at 45 FPS. 2. Project: "Deep-Path" (Medical Imaging) Tech: Custom Mask R-CNN architecture for gigapixel-scale histopathology segmentation. Impact: 0.94 Dice Coefficient; automated identification of 50k+ cells per slide, reducing manual review by 70%. 3. Project: "Flux-Analytics" (Retail Intelligence) Tech: 3D Pose Estimation using HRNet and temporal mapping for behavior recognition. Impact: 95% gesture recognition accuracy, providing automated "intent-to-buy" metrics for retail pilots. Core Competencies Architectures: YOLO (v8-v11), Vision Transformers (ViT), Mask R-CNN. Frameworks: PyTorch, TensorFlow, OpenCV, MediaPipe. Deployment: TensorRT, ONNX, Dockerized Cloud APIs, Edge IoT.