Ankit Kumar Singh Email & Phone Number

Bengaluru, Karnataka, India

MLOps Platform Development:Joined a key project focused on developing an MLOps platform on the Openshift cloud platform with Kubernetes. This end-to-end platform provides critical capabilities, including data acquisition, code development, model training, model tracking, deployment, and monitoring of services in the cloud. Key tools and frameworks used include MLflow for model tracking, Seldon Core and FastAPI for scalable model services, Jupyter and VS Code Remote for development, and Grafana and Prometheus for monitoring and observability. Integrated with CICD workflows through Tekton and ArgoCD, the platform seamless automation and scheduling, while also incorporating tools such as S3, ELK, and Ambassador for gateway and user authentication services. This work has enabled me to build solid expertise in Kubernetes and microservices.Data Engineering Platform Development:Currently working on building a dedicated data engineering platform to streamline data processes across the company, enhancing data accessibility, processing efficiency, and infrastructure scalability Documentation & CI/CD:Developed a documentation service and CI/CD pipeline using Docusaurus to streamline knowledge sharing and maintain consistent documentation practices across teams.

Bangalore Urban, Karnataka, India

MLFlow Integration & Team Training:Introduced MLFlow to streamline machine learning workflows, enhancing reproducibility and tracking across projects. Conducted training sessions and presentations to enable other data scientists to efficiently adopt the MLFlow platform in their work.Defect quantification with Synthetic Data Generation:Introduced an innovative approach to address a lack of annotated data in defect detection by using synthetic image generation techniques with OpenCV and Python. Developed a custom dataset to train a YOLO model for counting nodules and bubbles in semiconductor chips. This solution achieved 87% accuracy, significantly improving upon previous unreliable methods that had 40% accuracy.Chip Defect Classification:Developed and deployed a computer vision model of efficientnet series for classifying chip defects. Worked with high-resolution nanometer-scale chip images to identify and categorize defect classes.Custom TensorFlow Activation Functions and learning rates:Contributed to developing customer learning rates and custom activation functions within the TensorFlow package, tailoring it for computer vision applications in chip manufacturing.

Pune Area, India

Document Data ExtractionCreated a table extraction algorithm which Improved the API response time by (1.5 minutes to 19 seconds). The same algorithm helped extracting bank statements of SBI, ICICI, HDFC and other cooperative banks along with metadata of the client. (Tesseract-OCR, Faster-RCNN (field detection), Resnet (document classification), flask, rabbitmq, AWS-ec2)Facial AuthenticationIntegrated light weight face detector and encoder for low end devices (RabbitMQ, opencv, dlib, django, aws). Changed the classifier to KNN in place of SVM for low latency. Use cases where our project was deployed--Realtime monitoring employees working in office. Along with office entry authentication-Realtime monitoring employees working from home. (detecting person holding mobile, camera tampering, only the authenticatedperson sitting next to PC)Intrusion Detection in Indonesia MinesDeveloped a Computer Vision based POC for Intrusion of cars in specific region in mines field while detecting other classes of vehicles which are Trucks and Cranes. My Solution: Fine-tuned the YOLO-V3 model on personally acquired dataset of cars, trucks and cranes in different scenario. Implemented centroid tracker on car objects to detect intrusion in specific area.Retail Analytics in MallWorked on end-to-end development of the Project in which flow was to Detect and Recognize peopleentering and exiting the premises of Shopping Mall followed by Age, Gender, Emotion and dwell time of people. All the data generated is then visualized on the Dashboard.Technologies used: Computer Vision, Deep Learning, model server API development Tools used: Python, OpenCV, Tensorflow Keras, Django, Flask, Apache Superset (like Tableau), Redis, sqlite3, AWS (for model server API deployment), NVIDIA Jetson-NANO (for faster inferencing), Raspberry pi 4B

Pune, Maharashtra, India

Developed computer vision models for intrusion detection and retail analytics. Integrated model in the project for deployment. Developed superset Dashboard of Retail Analytics as per client requirements.

Bachelor Of Engineering - Be, Computer Science Engineering

High School, Science