Prathamesh Dinkar
Results-driven Machine Learning Engineer with 6+ years of experience transforming complex data into actionable business solutions. Expert in leveraging data science, machine learning, and deep learning to solve critical business challenges and drive innovation.
- Role
Senior AI engineer
- Years of Experience
7 years
- Professional Portfolio
View here
Skillsets
- TensorFlow - 4 Years
- PyTorch - 0.6 Years
- ML - 6 Years
- Deep Learning - 6 Years
- Artificial Intelligence - 6 Years
- Python - 6 Years
- Gen AI - 4.5 Years
- Data Visualization - 6 Years
- LLM - 2 Years
- Docker - 4 Years
- Quantitative Analysis - 4 Years
- Algorithms - 6 Years
- OpenCV - 3 Years
- Git - 3 Years
- anomaly detection - 2 Years
- AWS - 2 Years
- NLP - 2 Years
- Team Handling - 1.5 Years
Professional Summary
7Years
- Jul, 2022 - Present3 yr 2 months
Senior AI Engineer
Ignitarium Technology Solutions Private Limited - Jul, 2018 - Jul, 20224 yr
AI Engineer
Larsen & Toubro Limited
Applications & Tools Known
Tensorflow
PyTorch
.png)
Docker
scikit-learn
Ubuntu
Python
Project Jupyter
Anaconda
Keras
Scikit-learn
OpenCV
Git
Tableau
Power BI
plotly
Mediapipe
AWS
Tableau
Seaborn
Plotly
NLP
Huggingface
AWS
Work History
7Years
Senior AI Engineer
Ignitarium Technology Solutions Private LimitedJul, 2022 - Present3 yr 2 months
Spearheaded custom benchmarking initiatives for SOTA algorithms across classification, object detection, and segmentation, ensuring top-tier performance and accuracy. Engineered SOTA AI models for 3D segmentation of custom 3D characters, achieving an exceptional mIOU of 99%, setting a new standard in precision. Developed proof-of-concept (POC) solutions for AI-based motion capture and custom 3D character animation, demonstrating innovative approaches to dynamic character modeling.
AI Engineer
Larsen & Toubro LimitedJul, 2018 - Jul, 20224 yr
Implemented SOTA CNN models, including VGG16, ResNet50, and YOLO V3, to achieve high-performance benchmarks in various projects. Designed and built custom CNN models using TensorFlow and PyTorch, incorporating ensemble methods and transfer learning to boost model performance by 15%. Applied a range of machine learning algorithms, including regressions, tree-based models, and anomaly detection techniques, to address diverse analytical challenges. Executed anomaly detection algorithms and clustering techniques, successfully reducing Non-Revenue Water by 55%, showcasing significant operational impact.
Achievements
- Quarterly Award for Customer Satisfaction - Ignitarium Technology Solutions Private Limited
- Sport Award for Best Performance - Ignitarium Technology Solutions Private Limited
- Best Performer of the Year - Larsen & Toubro Limited
- Quarterly Award for Customer Satisfaction
- Sport Award for Best Performance
- Best Performer of the Year
Major Projects
8Projects
LLM Powered Video Search
Jul, 2022 - Present3 yr 2 months
Searching for specific video segments based on user query
Fine Tuning Video Foundation Model
Ignitarium Technology Solutions Private LimitedMay, 2024 - Jun, 2024 1 month
- Objective: To fine-tune the SOTA video foundation model for text-to-video retrieval task
- Solution: Implemented VAST model for fine-tuning for MSRVTT dataset for text-to-video retrieval
High-Resolution Image Generation for E-commerce
Ignitarium Technology Solutions Private LimitedJul, 2023 - Dec, 2023 5 months
To generate high-resolution product images from low-resolution inputs to enhance e-commerce visuals
Image Editing with GANs for Style Transfer
Jul, 2023 - Jul, 2023
To perform high-quality image editing by altering specific attributes while maintaining the overall content
Enhanced 3D Object Reconstruction from Images
Ignitarium Technology Solutions Private LimitedApr, 2023 - Mar, 2023
To create a detailed 3D model of an object from a 2D image
Out-of-Distribution Detection
Ignitarium Technology Solutions Private LimitedOct, 2022 - Jan, 2023 3 months
- Objective: To identify out-of-distribution (OOD) samples in classification, object detection, and segmentation task
- Solution: Implemented several OOD algorithms like MLS, MSP, ODIN, Energy, and Mahalanobis. Evaluated ResNet 50 for classification, SSD for object detection, and DeepLabv3+ for segmentation. For the classification task, AUROC is 99.8% and for the object detection task it was 92.5%
Pipeline Joints Detection using Deep Learning
Larsen & Toubro LimitedOct, 2021 - Mar, 2022 5 months
To detect and classify the pipeline fitting like bends, tee, valves, etc. from pipe network drawings
Detecting Leaks in the Water Supply Network
Larsen & Toubro LimitedMar, 2019 - Sep, 2019 6 months
- Objective: To detect leak in a water supply network using water consumption data from water meters
- Solution: Pre-processed the data from the water meters and created anomaly detection algorithms with PCA, Isolation Forest and KNN. This resulted in 55% reduction in Non-Revenue Water (NRW)
Education
Masters in Machine Learning and AI
Liverpool John Moores UniversityPost Graduate Diploma in Machine Learning and AI
IIIT, BangaloreB. Tech in Mechanical Engineering
Visvesvaraya National Institute Of Technology, Nagpur
Certifications
Deep learning specialization
Generative adversarial networks (gans) specialization
Generative ai: prompt engineering basics
Generative ai with large language models (llms)