Abhay Kumar
Bangalore, Karnataka, India
Email: abhay8280@gmail.com | Phone: +91-9731-812-812
LinkedIn Profile | Github Profile
About Me
I am a seasoned professional with over 10 years of experience in the fields of Machine Learning, Computer Vision, and Natural Language Processing (NLP). My career has been marked by delivering data products at scale, building successful Machine Learning teams from scratch, and contributing to project planning for data-driven products. I have a passion for solving complex problems and a track record of driving innovation in diverse domains, including retail, edtech, and media tech.
Domain Expertise
Skills:
- Computer Vision: Object Detection, Image Classification.
- NLP: Text Classification, Topic Modeling.
- Machine Learning: Deep Learning (PyTorch, TensorFlow), Scikit-learn.
- Data Processing: NumPy, Pandas, Spark, Hadoop.
- Cloud Platforms: AWS, Azure, Google Cloud.
- MLOps : Git, Docker, CI/CD, MLflow
- Team Building
- Team Leadership
- Data Product Development
- Project Planning
- Data Annotation Tool Design
- Data story telling and effective presentations
- Model training, optimisation and deployment at scale
- Stakeholders/client engagement
- Algorithm research and selection
- Technical to Business interpretation and vice versa
Professional Experience
Senior Machine Learning Engineer | Target Corporation
Feb 2021 - Present
- Led the development of cutting-edge computer vision algorithms that significantly improved product recognition accuracy for a leading e-commerce platform, resulting in a 20% increase in sales.
- Pioneered the implementation of NLP techniques to enhance the company’s chatbot, leading to a 30% reduction in customer support response times.
- Built and mentored a high-performing Machine Learning team from scratch, fostering a culture of innovation and excellence.
- Collaborated with cross-functional teams to define project scopes, timelines, and deliverables, ensuring alignment with business objectives.
Sr Data Scientist | Embibe
May 2019 - Jan 2021
- Managed the end-to-end development of Question tagging system.
- Designed and implemented in-house annotation tools to generate high-quality labeled data, which powered multiple Machine Learning models across the organization.
Senior Data Scientist - ClipIndia
Sep 2018 – Apr 2019
- Implemented NSFW video classification, reducing inappropriate content by 28%.
- Designed video asset tagging system, increasing DAU by 3%.
Senior Data Scientist - Cogknit Semantics
Sep 2015 – Sep 2018
- Built Vision Intelligence team from scratch and handled client projects
- Spearheaded catalogue filtration project, reducing incorrect product placements by 12%.
Sr. Research Analyst - Edureka
Dec 2013 – Sep 2015
- Conducted extensive research on Machine Learning course content and collaborated closely with Subject Matter Experts (SMEs) to develop comprehensive technical content
- Contributed significantly to the community by authoring blog posts on advanced Machine Learning and Big Data topics.
- Led the initiative to create a comprehensive knowledge base of blogs and most asked technical queries, reducing the workload on the TA team and enhancing the overall efficiency of support and learning resources.
Projects
Target Corporation
Anomaly Detection System for Data Pipeline
- Developed an anomaly detection system to prevent corrupt images from entering the data pipeline.
- Identified and blocked 30% of anomalous images, resulting in substantial reductions in processing time
- Improved downstream algorithms, leading to an 8% enhancement in business metric.
Invoice Digitisation System
- Implemented OCR model fine-tuning to detect and recognise specific style of text used in invoices.
- Finetuned transformer based layout analysis model to locate table areas for text extraction.
- Finetuned table transformer architecture and identified and reconstructed tabular structures.
- Digitized table data into structured JSON format, preserving row and column information as keys.
- Extracted key attributes from invoices for image indexing and efficient searching.
Real-Time Safety Monitoring System
- Developed a real-time monitoring system using camera streams to detect individuals in unsafe zones and ensure adherence to safety protocols.
- Recorded safety violations with evidence, raising safety incidents for review and training.
- Extended monitoring capabilities to cover a wider area, overcoming resource and weather constraints.
- Enhanced safety and operational efficiency.
- Reduced safety violation events by 40%.
Privacy Blur Feature Implementation
- Developed a privacy blur feature for archived video streams.
- Employed lightweight linear object tracking(kalman Filter) and yolo object detectors to reduce compute needs.
- Mentored junior team members, fostering their growth and enhancing the overall team’s performance.
Embibe
Question Tagging and Topic Extraction System
- Collaborated with SME, Data engineers and annotator team to gather and curate high-quality datasets for training and validation purposes.
- Designed and implemented an automated system to tag questions with subjects and extract relevant topics from text.
- Fine tuned Resnet based architecture for visual feature extraction and used BERT model for text embeddings extraction
- Enhanced content organization and searchability in a repository of millions of question papers.
- Utilized natural language processing techniques for precise content discovery.
- Developed question tagging using both text and image content, enabling better content recommendation and search accuracy.
CLIP India
Content Enrichment Engine
- Implemented NSFW video classification for Content Moderation, maintaining a clean platform by automatically detecting obscene content, leading to a 28% reduction in inappropriate content.
- Designed a video asset tagging system with multiple attributes, improving content discoverability and feed ranking, resulting in a 3% increase in DAU.
Feel free to get in touch with me at abhay8280@gmail.com for collaboration opportunities or to discuss exciting projects in the fields of Machine Learning, Computer Vision, or NLP.
Thank you for visiting my portfolio!