Aug 2022 – Jan 2025 · 2 yrs 5 mos · Bengaluru, Karnataka, India

• Led 10 high-value AI projects from concept to deployment, ensuring on-time delivery and client satisfaction. Acted as Customer anchor streamling operations, enhancing issue resolution, standardizing RCA processes, and upgrading algorithms to drive retention and revenue. Managed the Detection team and mentored juniors in AI technologies and best practices.
• Implemented ASH algorithm for out-of-distribution detection , cutting tagging efforts by 50% and uncovering new brand patterns. Further enhanced the accuracy by integrating Gemini.
• Implemented zero-shot brand identification using CLIP by comparing image and text embeddings, eliminating the need for manual tagging and training time.
• Implemented LightGlue algorithm for feature matching between images to detect overlap areas, preventing product over-counting on shelves and improving Share of Shelf metrics. and inference by 2 times faster.
• Trained YOLOv8 models, boosting product detection mAP by 5% on shelves. Leveraged TIDE for performance evaluation, identifying 6 key error types affecting mAP.
• Designed an algorithm using Grounding DINO and FAST SAM to generate soft-tagged segmentation data, reducing tagging time by 50%.
• Leveraged VLMs like Gemini and ChatGPT for POSM identification, eliminating tagging and training time.
• Built a gold data sampling service to standardize session sampling, ensuring accurate evaluation of detectors, classifiers, and pipeline performance.

Sep 2020 – Aug 2022 · 1 yr 11 mos · Hyderabad, Telangana, India

• Built a multi label classifier , to classify palm fruits into different categories like grade and stalk. On the limited dataset got an accuracy of around 75 - 80%.
• Lip Sync : Trained a Conditional Gan (Face and Audio Encoder and Face Decoder), to generate the lips in sync with the target speaker. Two discriminators were used to keep check on the sync and the quality of the generated lips. Further Super Resolution was used to improve the quality of the lips.
• Video Analytics : Used Nvidia’s deepstream , to handle multiple CCTV feeds and run various models like mask detection, entry & exit count etc and simultaneously push the data to cloud using Kafka.

Jan 2020 – Aug 2020 · 7 mos · IIIT-H,Hyderabad

• Worked on sports visual tracking and action identification.
• Generated heatmaps for player movement on the court and identified shots and strokes using LSTM.
• Collaborated with Star Sports for live match analysis in Premier Badminton League (PBL) 2020 .

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

• Designed an efficient brand logo detection model using YOLO algorithm that could detect various brands appearing on Youtube videos.
• Worked on project based on ad targeting which consists of classifying Youtube videos as adult or safe for playing ads on the basis of thumbnails using neural networks like VGG 16 and Resnet-50.

May 2018 – Jul 2018 · 2 mos · India

• Part of summer workshop on machine learning and deep learning in computer vision.
• Built a CNN model that was capable to recognize handwritten text.