Discover how our engineering expertise helps businesses secure infrastructure, prevent data leaks, and stay ahead of cyber threats with scalable, real-world solutions.
Exploratory data analysis performed on file to identify the features and used embedding models to generate vector features.
URL's are segmented as parameters, domain, etc. and processed using TF-IDF, BoW, Contextual Embedding for further classifying as a malicious or benign URL.
Random forest, LGBM models are used to scan the executable file on all platform. File headers and file entropy based features are consumed for detection. FP file analysis and mitigation done using shape values of file.
Data from files (.docx, .pdf) is classified using LLM models like Mistral, Llama. Fine-tuning techniques like LoRa, SFT, PEFT were used on Bert, Mistral models to tune the model based on industry categories like Financial, Legal, etc.
PE files are converted to image and analysis is performed using deep learning models like ResNet, LeNet and AlexNet on .text section of files.