Ai Developer

Automated resolution of customer limit increase requests, improving operational efficiency during on-call rotations. Enhanced Robot Detection features by making critical headers available in the Nginx release, ensuring accurate classification of traffic types. Received a full-time offer for continued impact and dedication within the same team.

Responsibilities include handling discussion sessions, maintaining weekly office hours to assist students.

As a researcher, I was involved in designing, testing, network optimisation and deployment of state-of-the-art computer vision detection, redaction, and recognition algorithms for real-time inference on surveillance video data. Developed a joint learning lightweight CNN for facial landmarks and attribute prediction. The model uses a SpatialCord Conv layer which exploits spatial coordinate information (which CNN cannot extract) that leads to an accurate landmark detection (0.041 NME on WFLW). Reduced the inference time of the company’s existing facial recognition application by 31 folds on CPU/GPU machines, saving both time and money by optimizing network architecture. Led a group of 8 Interns in producing an efficient marine animal tracking system for marine biologists to observe behavioural and growth statistics of a species under a constrained environment. Our framework performs reasonably well in real-world environments. Implemented state-of-the-art face/object detection (MTCNN, SSD, RetinaNet, EfficientDet, Faster RCNN, and CenterFace) and recognition algorithms from scratch to reproduce the published results.

Worked on extracting geological images from a group of images that comprises tables, geological maps, grids, and drill photos in an unsupervised manner. For a mining corporation, we designed a real-time system for geo-tagging a mining document to a specific location and extracting ppm of minerals that slashed the manual work by 80%.