Anshu Garg

I deliver data-driven insights and production-ready analytics solutions to improve manufacturing processes and product quality. I work closely with stakeholders across departments to translate business problems into high-value data/AI use cases and implement scalable solutions end-to-end. Current focus: Building a centralized, enterprise data platform for ifm subsidiaries to enable standardized data foundations and consistent, cross-site analytics—using Azure Data Warehouse technologies and modern cloud engineering practices. Key responsibilities & impact 1. Centralized Data Platform: Designing and developing a centralized enterprise cloud data platform to enable standardized data foundations and reusable analytics across ifm subsidiaries. 2. Automated SPC Solution: Delivered a fully automated Statistical Process Control (SPC) solution that identifies consistently out-of-control process steps and supports targeted corrective actions to improve production quality and operational efficiency. 3. End-to-End Azure ETL Pipelines: Built robust ETL pipelines to ingest, transform, and integrate distributed production data from raw text/XML result files and SQL databases. 4. Azure Analytics Stack: Leveraged Azure Synapse Analytics (Dedicated SQL Pool), Azure Data Lake Gen2, and Power BI to deliver scalable data warehousing and analytics solutions. 5. CI/CD & Engineering Practices: Implemented CI/CD pipelines using Azure DevOps, improving deployment reliability, reproducibility, and development velocity. 6. Shopfloor AI & GenAI: Introduced machine learning use cases for production process optimization and developed an agentic GenAI chatbot to enable faster access to shopfloor production insights. 7. Workflow Automation: Automated manual processes and improved data accessibility using Power Automate and Power Apps.

Contributed to an EU-funded research project focused on 3D hand pose estimation and gesture recognition using stereo RGBD video captured via Intel RealSense–type depth cameras. Built and evaluated state-of-the-art deep learning pipelines for accurate, real-time 3D hand localization and tracking, combining research rigor with production-grade engineering. Key contributions: 1. Designed an end-to-end 3D hand pose estimation pipeline using SOTA deep learning approaches on stereo RGBD frames, targeting robustness under real-world noise and occlusions. 2. Developed a novel depth filtering technique to reduce depth noise and improve hand localization accuracy in depth-based processing. 3. Implemented a robust one-to-one hand mapping approach across stereo frames to ensure temporal and spatial consistency. 4. Performed stereo camera calibration and built comprehensive data preprocessing workflows to generate accurate model inputs. 5. Engineered a high-performance C++ real-time capture module for stereo camera streaming with zero frame drops, enabling reliable online inference workflows. 6. Built a Python interface for seamless data retrieval and integration from ROS2 nodes, supporting modular experimentation. 7. Containerized and deployed the full solution in a Docker environment with ROS2, enabling portable execution, reproducible experiments, and easy integration with robotics middleware. 8. Conducted extensive literature reviews, reproduced and benchmarked multiple SOTA models, and iterated systematically to improve accuracy and stability.

Designed and prototyped machine learning approaches to support safer AI in autonomous driving, focusing on detecting dataset distribution shifts and improving robustness under out-of-domain (OOD) conditions. Key contributions: 1. Designed ML applications for distribution shift detection to identify when models encounter data that differs from training conditions (safety-critical robustness). 2. Built an end-to-end ML pipeline PoC for experimentation and reproducibility using MLflow, Optuna, PyTorch, and PyTorch Lightning (tracking, tuning, repeatable training runs). 3. Generated OOD samples by exploring and sampling VAE latent spaces, enabling controlled evaluation of model behavior beyond in-domain data. 4. Visualized high-dimensional in-domain vs OOD data using t-SNE, UMAP, and other dimensionality-reduction techniques to interpret separation, clusters, and shift patterns. 5. Implemented and compared multiple approaches for classification and anomaly detection on both in-domain and OOD datasets, analyzing model sensitivity and failure modes. 6. Supported research execution through literature review, experiment design, and clear documentation of findings and trade-offs.

I recently joined TUK University to pursue my Masters in Computer Science. I am highly interested in learning Machine Learning and Artificial Intelligence courses. My interest inclines towards specialization in Intelligent Systems. I have recently completed A1 level German course and will start learning A2 level soon.

Object Detection Using Transformer Fusion and Detection Transformer on Multi-Sensor Data for Automated Driving - designed a novel transformer-based architecture TransFuserDeTr model for 2D Object Detection using RGB image and BEV image for Kitti and NuScenes dataset.

1. Designed Machine Learning applications for enhancing crop production by predicting flowering days using multi-modality data (pictures, weather text data etc). 2. Data imputation for weather data. 3. Designed new optimized technique to represent one-hot encoding to store DNA data.

Researched and designed a new project architecture pipeline framework using PyTorch Lighting and MLFlow. Pre-processed data and used different training methods on Variational AutoEncoder. Used CUDA and SLURM for training ML models.

I worked on the LifeCare product of Tieto. I mainly had the responsibility to migrate the functionalities of there legacy system to the new technology stack. Here I worked on the following technologies - C#, MVC, Web Services, Factory Design Pattern, Javascript, Knockout JS, Web API. I worked with a Swedish client and learned a lot about their business requirements and way of working of the municipality of Sweden.

Working as a valuable employee of Infosys is great experience. Every day I try to learn something new which also benefits to company and society indirectly. I worked as the Full Stack .NET Developer. I had a very rich experience in C#, Javascript, HTML, SQL Server Management Studio, deployment processes, communication with international clients.

I joined Infosys as an intern and completed training. Worked as SGT for my batch. Training shaped me to an employee from a student and learned professional attitude.

Student life is best time to learn and expand in all the dimensions and then decide which field amaze you the most. Learning and fun should be balanced, both are equally important and friends make your journey amazing and memorable.