Apr 2024 – Jan 2025 · 9 mos · On-site

• Deco: AI-Powered Home Interior Design App
• Led design, development, and launch of Deco App’s AI-powered features, achieving 1,000+ DAU in North America
• Created and deployed APIs for Style Transfers and Image Inpainting (Diffusion Models) using FastAPI in Python
• Innovated a pattern printing model for Wall and Floor using a single camera image, with a scale error rate of < 3%
• Engineered model for Empty Room Virtual Staging using segmentation mask, depth mask, and Stable Diffusion

May 2021 – Apr 2024 · 2 yrs 11 mos · On-site

• Virtual Try-On [VTO]
• Designed and trained model for real-time ear keypoint detection on edge devices using MobileNet with MAE 1.2
• Decreased 8X delivery time of VTO web app. by designing and building an SDK to track the key points on the face
• Optimized face smoothening filter using the Bilateral Filter, boosting the performance by over 200% to 45 FPS
• Trained a Segmentation model for wrinkle segmentation used in a skin care app with an accuracy of more than 99%
• Enhanced user experience of VTO app. by stabilizing real-time face landmarks pts. of the face using One Euro Filter
• Avatar Creation(Selfie to Avatar)
• Created image classification model to classify a person’s characteristics i.e hair and beard type, glasses, and gender
• Developed and deployed a face texture extraction algorithm which extracts face texture from a given selfie image

May 2020 – Sep 2020 · 4 mos · Bengaluru, Karnataka, India

• Project Title: Digitisation of complex Engineering Drawings using Computer Vision and Deep Learning
• Applied Connected Component Algorithm to separate texts and noises from the given Engineering Drawing image
• Recognized and eliminated the connecting wires based on Hough Line Transform and length threshold respectively
• Localized the symbol's position in the image by Contour Detection followed by Contour Merging based on proximity
• Trained CNN model to classify 140 classes of symbol-using resnet-50 architecture and achieved an accuracy of 97%
• Created a NetworkX graph (stored in XML) based on connection information of symbol for the aid of visualization

Dec 2019 – Feb 2020 · 2 mos · Kharagpur Area, India

• Principal Instructor: Prof. Pranab Kumar Dutta, Department of Electrical Engineering, IIT Kharagpur
• Trained a CNN model on the CT scan images of lungs to predict if the lungs are benign or malignant in nature
• Used Otsu's algorithm to perform Image Segmentation on CT scan images of lungs to extract the Region of Interest
• Improved classification performance from the accuracy of 65% to 89% by transfer learning using VGG Architecture

May 2019 – Jul 2019 · 2 mos · Bengaluru Area, India

• Information Extraction using MITIE and Spacy:
• Developed a Spacy model to parse start date, end date, cost and other relevant data from a software licensed pdf
• Improved performance of the spacy model, by training another model using MIT Information Extraction software
• Software Testing and Data Retrieval using IoT:
• Worked on improving BangDB 2.0 Software by writing test cases for BangDB 2.0 in C and C++
• Used open source SWIG software to wrap BangBD 2.0's C++ code to make it compatible with Python platform
• Prepared a report on Data Retrieval of AC's temperature, current and voltage by using Arduino and IoT sensors