Education
B.E. in Information Technology
Pune Institute of Computer Technology, SPPUPICT is a well-known college in Pune. I completed my B.E. in Information Technology. During my tenure at PICT, I completed four internships in the field of ML and DL, published two international research papers and had been two times smart India hackathon finalist.
Diploma in Computer Engineering
Government Polytechnic Gondia, MSBTEGondia is a small town near Nagpur, I have spent my 3 years here learning the basics of computer science, Explored various domains such as web and android development, Cyber Security, UI/UX and game development. I also got the opportunity to lead various projects.
Experience
Monsoon CreditTech
Senior Data Scientist Aug 2023 - Present
- Leading multiple research initiatives focused on automating end-to-end ML model lifecycle, from deployment to monitoring and reporting.
- Spearheaded product evolution through strategic feature enhancements and innovative additions, while fostering seamless collaboration with cross-functional development teams.
- Architecting and implementing sophisticated ML solutions including fraud detection and risk assessment models, serving a diverse portfolio of banking clients.
- Successfully delivered risk and propensity modeling solutions for a major Indian bank, demonstrating strong stakeholder management and communication skills.
- Led comprehensive audits of critical ML risk models, ensuring alignment with regulatory standards, governance frameworks, and business objectives.
- Engineered robust API solutions for default prediction, integrating them with the core LOS system through a sophisticated tech stack of Python, distributed SQL, and R.
- Developed an innovative automated feature engineering framework that significantly enhanced model performance and reduced development time across projects.
- Fast-tracked a critical banking project, delivering comprehensive solutions with robust validation in half the estimated timeline while maintaining high quality standards.
- Contributed to the design and implementation of scalable ML infrastructure, focusing on maintainability and reusability principles.
- Developed models for loan delinquency prediction, working with diverse data sources from banks and NBFCs to improve credit risk assessment.
PMV Electric
R&D Intern Computer Vision
- Worked on affordable Level 3 autonomy, self-driving cars
- Developed a light, Novel navigation which uses Semantic Segmentation and SLAM for localization
- Created a system which cost 70% less than the previous System
AP Analytica
Deep Learning Intern
- Deployed image-based invoice parsing systems
- Perform a Rigorous Benchmarking of Various models and employed multiprocessing techniques to enhance performance by 60%
- Deployed the system as a web-application using flask
ALIVETECH
Android App Developer
- Developed an Event Management App with Firebase as Database
- Another App was Number Finding App, Which assisted the Traffic police in scanning the number plate
Publications
CascadeTabNet: An approach for end to end table detection and structure recognition from image-based documents
Devashish Prasad, Ayan Gadpal, Kshitij Kapadni, Manish Visave, Kavita Sultanpure.
CascadTabNet is an automatic table recognition method for interpretation of tabular data in document images. We present an improved deep learning-based end to end approach for solving both problems of table detection and structure recognition using a single Convolution Neural Network (CNN) model. CascadeTabNet is a Cascade mask Region-based CNN High-Resolution Network (Cascade mask R-CNN HRNet) based model that detects the regions of tables and recognizes the structural body cells from the detected tables at the same time. We evaluate our results on ICDAR 2013, ICDAR 2019 and TableBank public datasets. We achieved 3rd rank in ICDAR 2019 post-competition results for table detection while attaining the best accuracy results for the ICDAR 2013 and TableBank dataset. We also attain the highest accuracy results on the ICDAR 2019 table structure recognition dataset
Pre-print More and CodeHOG, LBP and SVM based Traffic Density Estimation at Intersection
Devashish Prasad and Kshitij Kapadni and Ayan Gadpal and Manish Visave and K. SultanpureWe propose an efficient way to estimate the traffic density on intersection using image processing and machine learning techniques in real time. The proposed methodology takes pictures of traffic at junction to estimate the traffic density. We use Histogram of Oriented Gradients (HOG), Local Binary Patterns (LBP) and Support Vector Machine (SVM) based approach for traffic density estimation. The strategy is computationally inexpensive and might run efficiently on raspberry pi board
Pre-print More and CodeProjects
Dynamic Human Authentication Using Videos
Smart India Hackathon 2020, DRDODeep-learning, Computer-Vision and System Design
- Work on Human Authentication project using Computer Vision, multiprocessing and Django
- Developed Real-time Human Authentication using camera video of Gate entrance with dynamic human registration
- Built and integrated state of the art algorithms for Face Detection, Recognition with Gait Recognition and Pose Estimation
- To enhance security, We also added anti-spoofing
- Deployed the system on Django and employed multiprocessing to improve performance significantly
Faculty-Student Feedback Management System
PICTMySQL, JSP and System Design
- JSP and MySQL based online system to record feedback of students of all departments of PICT.
- Report generation and visualization of feedback data
Automatic 3D Object Inspection System
Consultancy project, VEMDeep-learning, Computer-Vision and System Design
- Automatic system for dimension extraction of 3D objects for quality inspection
- Improved accuracy significantly over previous systems and reduced manual work by automating the process
- High tolerance and accuracy based object dimension extraction system using cameras and FAIR report generation. 3D reconstruction using multiple cameras Alice Vision API.
Smart Traffic Junction
Paskathon (PICT ACM hackathon) and Personal ResearchDeep-learning, Computer-Vision and System Design
- The main objective of this project was to adjust the traffic light and there on off time based on traffic and to provide an interface for traffic management to record and control all traffic junction live feed
- Used HOG + LBP and SVM based Machine Learning approach to make it deployable with limited hardware
Online Inspection of Packed Tobacco Cases
Smart India Hackathon 2019, ITCDeep-learning, Computer-Vision and System Design
- Machine Learning based real time Tobacco leaves grade classification based on color and texture.
- Unsupervised learning techniques like PCA and LDA. Feature extraction using Gabor, Haralick, LBP and HOG
- Deployed Using QtPy
Event Management Android Application With Firebase
Pict Csi Student BranchApp Development, Database Schema Design and System Design
Android App for registration of participants in Enthusia Event
Vehicle Number Finder App
Pict, Audit courseComputer Vision, Android and System Design
Scans the vehicle number plate, and extracts information about the owner of the vehicle
Skills
Artificial Intelligence
Machine Learning • Data Science
Deep Learning • Computer Vision
Natural Language Processing
Generative AI
Software Development & Architecture
Software Configuration Management
Requirements Management
System Architecture Design
Software Development Life Cycle
Deployment & Infrastructure
CI/CD Pipeline Management
Cloud Infrastructure
Scalable Architecture
System Integration
Management & Leadership
Team Leadership
Project Management
Risk Mitigation
Stakeholder Communication
Languages
Python
Java
R
C++
Frameworks & Libraries
PyTorch
TensorFlow
OpenCV
NumPy & Pandas
Matplotlib & Seaborn
Scikit-learn
Hugging Face
PySpark
Development & DevOps Tools
Docker
Git
Jira
Jenkins
Kubernetes
AWS
Databases & Data Warehouses
MySQL
PostgreSQL
MongoDB
Vertica
Impala
Firebase
Internet of Things (IoT)
Raspberry Pi
Arduino
Web Development and Designing
JS, HTML and CSS
Photoshop and Illustrator
JSP and PHP
Open Source Projects
CascadeTabNet 1.5k+ Stars 430+ Forks
Core ContributorAn end-to-end table detection and structure recognition approach for image-based documents. Published at CVPR 2020.
TextTron 50+ Stars 4 Forks
Creator & MaintainerA lightweight text detector for document images using image processing techniques. Designed for fast text detection without ML/DL models.
Document Image Augmentation 32+ Stars 2 Forks
Creator & MaintainerA comprehensive tool for augmenting axis-aligned document images, supporting various augmentation techniques including dilation, smudge, and color transformations.
E-Annapurna 13 Stars 6 Forks
Creator & MaintainerA platform connecting restaurants with people in need, facilitating food sharing and reducing waste. Includes both website and management panel.
