Arbaz Aslam

Data Science and AI Assistant

NovaTech • Temps plein

Sep 2025 - Mar 2026 • 6 mos

Enterprise Workflow Automation & Multi-Agent AI Architecture | Novatech During my tenure as a Data Science and AI Assistant at Novatech, I engineered an end-to-end data intelligence and workflow automation ecosystem designed to eliminate manual operational overhead, consolidate workforce analytics, and deploy autonomous AI systems. The initiative centered on bridging disparate business tools through automated data pipelines, architecting collaborative agentic workflows, and providing executive leadership with real-time project visibility. To establish a unified data layer, I built automated ETL pipelines using n8n to continuously ingest project and workforce records from multiple sources. I developed custom Python scripts for data cleaning, validation, and transformation to enforce strict data integrity and standardization across all workflows. This reliable data foundation fed directly into a centralized executive dashboard that provided granular KPIs, milestone timelines, and workforce performance metrics for high-level decision-making. On the AI engineering front, I architected a collaborative multi-agent system using CrewAI, establishing autonomous role-specific agents—Researcher, Analyst, and Writer—to handle complex research and synthesis workflows. Alongside this, I built an LLM-orchestrated project tracking bot to automate progress updates and status monitoring. To satisfy enterprise compliance and security standards, I deployed local-first conversational AI models via Ollama, ensuring 100% data privacy and zero external data exposure for sensitive internal queries.

Data Scientist

10Pearls • Temps partiel

Mar 2025 - May 2025 • 2 mos

Karachi AQI Forecasting & Production ML Pipeline | 10Pearls An end-to-end predictive machine learning system designed and deployed during my tenure at 10Pearls to forecast the Air Quality Index (AQI) for Karachi, Pakistan. The project automated the entire data lifecycle—from ingestion and serverless feature storage to model training, experiment tracking, and real-time interactive dashboard deployment. Key Metrics & Technical Highlights Predictive Accuracy: Achieved an R^2 score of 0.80 using an optimized XGBoost regression model trained on multi-variate atmospheric and meteorological parameters. Forecast Horizon: Delivered 72-hour (3-day) rolling AQI forecasts updated continuously via scheduled data pipelines. Feature Pipeline: Built automated data ingestion workflows integrating external historical and real-time weather APIs. Data Storage: Configured Google BigQuery as a scalable, serverless feature store for high-throughput feature retrieval and transformation. MLOps & Tracking: Managed the full model lifecycle including parameter tracking, metric logging, and artifact versioning using MLflow. CI/CD Automation: Built automated GitHub Actions pipelines for continuous integration, model retraining triggers, and automated application deployment. Dashboard Serving: Designed and deployed a responsive Streamlit web application to visualize live air quality trends, risk categories, and multi-day projections. Core Technical Stack Language & Core Libraries: Python, Pandas, NumPy, Scikit-learn, XGBoost Data Engineering & Storage: Google BigQuery, REST APIs MLOps & CI/CD: MLflow, GitHub Actions, Git Application & Deployment: Streamlit, Cloud Hosting