Ali AminiManesh

Conducting cutting-edge research on Unsupervised Anomaly Detection using Vision-Language Models at the Computer Engineering Department.

• Conducting novel research on Unsupervised Anomaly Detection (UAD) using Vision-Language Models (VLMs), specifically targeting logical anomalies in complex datasets (e.g., LOCO).
• Designing and implementing advanced prompt engineering techniques to enhance model performance in zero-shot and few-shot settings, reducing dependency on large-scale labeled data.
• Improving the accuracy of SOTA architectures by integrating custom attention blocks and refining feature extraction layers to better capture subtle visual inconsistencies.
• Co-authored a technical proposal for the TUBITAK 1001 Scientific and Technological Research Projects Support Program, focusing on next-generation anomaly detection systems.

Delivered end-to-end ML solutions for Iran's largest e-commerce platform, covering forecasting, segmentation, pricing, and business intelligence.

• Engineered a high-precision daily order forecasting model integrated with Explainable AI (SHAP, ACE), improving prediction accuracy by 25% and enabling proactive inventory optimization.
• Applied unsupervised K-Means clustering to detect anomalous reseller behaviors, enabling the B2B team to customize CRM strategies and increase targeted campaign conversion by 20%.
• Developed a dynamic discount recommendation engine based on profitability patterns, automating pricing strategies to boost average sales margins by 7%.
• Designed and deployed interactive Streamlit dashboards with Plotly, providing real-time KPI monitoring and reducing manual reporting overhead by 65%.

Drove data-informed decision making through statistical modeling, dashboard development, and automated data pipelines.

• Leveraged the PyData stack (Pandas, NumPy) and IBM SPSS to clean and model complex datasets, uncovering statistical trends that drove strategic business decisions.
• Designed and maintained interactive dashboards in Looker Studio and Metabase to visualize performance metrics, directly supporting cost reduction initiatives.
• Engineered automated data pipelines to streamline processing workflows, resulting in significant improvements in time efficiency and operational data accuracy.

TUBITAK 1001 — Next-Generation Anomaly Detection

Co-authored a scientific research grant proposal for the TUBITAK 1001 Scientific and Technological Research Projects Support Program, focusing on advancing anomaly detection systems using Vision-Language Models.

• Contributed to research design, methodology, and technical writing of the grant proposal.
• Targeted logical anomaly detection in industrial datasets using few-shot and zero-shot VLM approaches.

Mentoring undergraduate students on research methodology and machine learning fundamentals.

Taught Python programming and data science fundamentals to aspiring developers and analysts.