Prislo is an AI-powered shopping search engine that has 40 million products from 6k online stores in 6 countries (US,UK,PK,IN,CA,AU).- Developed COEUS (Context optimized embeddings for Universal Shopping), a language model that is domain adopted on 40 million products, fine tuned on 8 million products and 2.5 million shopping queries.- Empowered Prislo by Neural Search using COUES embeddings and HNSW graphs for Approximate Nearest Neighbor- Designed and developed ATHENA, an unsupervised AI trained on 5.2 million online shopping products. ATHENA clusters the products and generates new shopping categories by self analysis.- Designed and developed MORPHEUS, an autonomous web crawler that self-learns structures of e-commerce websites and extract products' data. It is capable of scraping thousands of websites in parallel.- Developed Prislo's search algorithm. It estimates product prices and incorporate user behaviors to understand the shopping intent. It outperforms many world-known shopping search engines in the quality and accuracy of search results.

TPL Maps is Pakistan’s first indigenously developed digital maps. Produced some cutting edge research and developed following applications.- Multi platform 3d map rendering engine- 360 panoramic image stitching algorithm- TPL Street Vision- Infrastructure Development and Demographics estimation from satellite imagery of Pakistan- Urdu Speech Recognition based on Kaldi- Urdu Speech Synthesizer- AI based Point of Interest mapping from 360 video survey of Pakistan- Drone flight planning to reconstruct 3d structures from multi perspective images- 3d point cloud extraction from multi-view imagery based on a technique called Structure From Motion.

Worked on state-of-the-art research in biomedical signal processing and machine learningResearch Title:Analysis of Human Emotions using advanced Signal Processing Algorithms on EEG dataDeveloped some techniques for :- Statistical analysis of brain signals- Feature extraction from brain signals using Multivariate Empirical Mode Decomposition (MEMD)- Multiple Classification models for 6 basic human emotions using regression and Artificial Neural Networks

- Developed a cutting edge jpeg compression algorithm for PhotoShrinkr, it maintains finest visual quality by applying different DCT encoders for different types of visual details in an image and it achieves high compression in ways that other apps do not.- Steganographic watermarking for iWatermark app. This technique is used to embed secure watermark in an image.