
Edula Vinay Kumar Reddy
Master's in Computer Science
Indian Institute of Technology Kanpur
About Me
I'll ask Google's Gemini AI instead 😉: Click here
Professional Experience
- Developed REST APIs for a Document Processing platform using Spring Boot in a microservices environment.
- Implemented login page authentication functionalities, custom password encryption mechanism, and email notification backend services.
- Developed an AI-based device to reduce surgical errors by validating patient wristbands during preparation and storing results in the electronic medical record via API.
- Implemented using image processing, YOLOv8, Mediapipe, and Raspberry Pi 4B; successfully tested at the Madurai branch hospital.
- Performed web scraping to extract data on medicines and medical services using Selenium and BeautifulSoup.
- Created a database to store scraped data using MongoDB in AWS Cloud.
- Published 57 articles and improved nearly 1,450 articles on topics including Python, Java, Machine Learning, Computer Networks, and Web Mining.
Education
Relevant Coursework:
- ML Pool:
- CS771 -- Introduction to Machine Learning (Dr. Piyush Rai) -- 10/10
- CS772 -- Probabilistic Machine Learning (Dr. Piyush Rai) -- 10/10
- CS779 -- Statistical Natural Language Processing (Dr. Ashutosh Modi) -- 10/10
- CS661 -- Big Data Visual Analytics (Dr. Soumya Dutta) -- 10/10
- CS776 -- Deep Learning for Computer Vision (Dr. Priyanka Bagade) -- 10/10
- Theory Pool:
- CS663 -- Computational Geometry (Dr. Sanjeev Saxena) -- 9/10
- Systems Pool:
- CS724 -- Sensing Communications and Networking (Dr. Amitangshu Pal) -- 9/10
Achievements
Secured the second-highest CGPA in the CSE department.
1st place in both Machine Translation and Sentiment Classification Competitions of CS779 course (Statistical NLP), IIT Kanpur, 2024.
National Finalist (Top 10) in Idea Summit 2021.
Projects
- A printed circuit board (PCB) is a thin board that electrically connects and supports electronic components using conductive tracks.
- My task is to Researching/Developing an end-to-end DL pipeline to automatically detect and localize surface defects on PCBs.
- Fine-tuning existing object detection models and integrating attention modules to improve sensitivity to small defects in real time.
- Built an image classifier using CLIP embeddings and MLPs to organize a user’s wardrobe images into categories and attributes.
- Implemented text-based occasion detection to map user input (e.g., “wedding”, “casual”) to predefined outfit categories.
- Designed a 16-head Transformer model that learns compatibility across FashionCLIP embeddings with category tokens, supporting compatibility prediction and complementary item retrieval.
- Evaluated on the Polyvore dataset, outperforming several state-of-the-art baselines on key metrics for outfit compatibility and retrieval.
- Developed a user interface (Image Selector, Predict Attributes, Full Outfit Generator, Occasion Outfit, Delete Image, User Mode) to upload, tag, and generate complete or occasion-specific outfits with one click.
- Improved the original RHO-LOSS framework described in “Prioritized Training on Points that are Learnable, Worth Learning, and Not Yet Learnt” .
- Replicated the RHO-LOSS baseline on CIFAR-10, matching the seed paper’s reported evaluation metrics.
- Proposed methods: Diversity-Augmented Batch Selection (DivBS) + RHO-LOSS, Exponential Moving Average–Adaptive RHO-LOSS (EMA-Adaptive RHO), Entropy-Regularized RHO-LOSS, RHO-LOSS + Bayesian Active Learning by Disagreement (BALD), RHO-LOSS + Margin Sampling, and Hybrid RHO-LOSS–Diversity Selection (RHO-DCAST).
- Performance highlight: DivBS + RHO-LOSS and RHO-DCAST outperformed the seed RHO-LOSS model in training speed, while EMA-Adaptive RHO outperformed on data efficiency (2.8% fewer samples to reach ~80% by epoch 50).
- Developed an English-to-Hindi/Bengali translation system that secured 1st rank in the IITK CS779 competition.
- Experimented with encoder-decoder architectures, including RNN-based models and attention mechanisms, to improve alignment.
- Implemented a Transformer-based model with pre-trained GloVe embeddings and language-specific preprocessing, optimizing training with mixed precision for 10× speedup.
- Created a majority-voting ensemble combining RNN with Focal Loss, RCAN, and BiLSTM (with Global Max Pooling) to tackle class imbalance.
- Engineered custom embeddings by integrating POS tags, dependency features, negation handling, and a word-level positivity score.
- Achieved 1st place with a 69.7 F₁-score in the IITK CS779 sentiment classification competition.
- Created personalized workout plans using GPT-3.5, collecting user preferences via a Tkinter GUI.
- Integrated Mediapipe for real-time exercise tracking, providing posture feedback through body landmark analysis.
- Used Vosk for speech recognition to switch exercises and reset progress hands-free.
- Developed a performance dashboard with Dash and Plotly, storing metrics in SQLite to visualize user progress over time.
- Automated rider detection to identify two-wheelers on the road.
- Implemented helmet detection using YOLOv8 to flag violations in real-time.
- Extracted license plate numbers with Tesseract-OCR and integrated the data into a backend database for fine processing.
- Developed a web application using Microsoft Azure Cognitive Services to power a chatbot for COVID-19 queries.
- Fetched and filtered live tweets (by city and resource type) to help users find nearby COVID-19 relief services.
- Displayed real-time statistics on total confirmed cases, recoveries, and death tolls globally and by country.
- Developed a smart irrigation system using IoT sensors (soil moisture, temperature, humidity) to collect field data and store it in the cloud.
- Applied ensemble learning techniques to predict rainfall patterns and optimize irrigation schedules, reducing overall water consumption.
- Automated water pump control based on real-time sensor inputs and rainfall forecasts to improve agricultural efficiency.
- Created a web-based platform for VIT hostel students to report and track complaints across categories like WiFi, hygiene, mess, security, and facilities.
- Implemented four user modules: students (register & track complaints), hostel staff (resolve facility issues), mess staff (resolve mess-related complaints), and VHS admin (manage overall ticket flow).
- Streamlined complaint categorization and status updates, improving resolution efficiency and transparency for all stakeholders.
Publications
Certifications
Technical Skills
- Programming Languages: Python, Java, C++
- ML Libraries: PyTorch, spaCy, scikit-learn, NumPy, Pandas, Ultralytics, LangChain, OpenCV
- Web Technologies: Spring Boot, HTML, CSS, PHP, JavaScript, jQuery
- Web Scraping: BeautifulSoup, Selenium
- Database Management: MySQL
- Developer Tools: GitHub, GitLab
- Cloud Technologies: Microsoft Azure, AWS
Leadership & Extracurricular
Published multiple high-quality articles on data science, AI, and machine learning topics as a member of the Analytics Vidhya Creators’ Club, contributing to a community of 3M+ users.
Collaborated with a global network of tech enthusiasts, enhancing skills in AI, cloud computing, and leadership; organized events and shared knowledge across the student ambassador community.
Participated in various NCC activities, demonstrating discipline and teamwork through successful completion of training and earning the ‘A’ certificate.