I build intelligent systems that sit at the intersection of applied machine learning, agentic AI, and cybersecurity — from training GNNs and autoencoders for intrusion detection to shipping full-stack ML applications that put those models to work.
- 🔭 Currently finalizing SentiNet, an adaptive AI-based network intrusion detection & prevention system, for my final year project
- 🧠 Focused on: agentic AI, graph neural networks, and LLM-based autonomous agents
- 🏆 Also competing on Kaggle in my spare time
- 📫 Reach me at kawish918@gmail.com
|
Adaptive AI-based network intrusion detection & prevention system — my final year project. A FastAPI central server, React admin dashboard, and Windows endpoint agent work together, backed by an Autoencoder/LSTM/GNN detection stack and a ReACT-based SOC agent running on Llama-3.1-8B-Instruct.
|
An edge-attributed graph attention network for intrusion detection that folds 40-dimensional flow features directly into attention and classification. Introduced a pragmatic class-filtering strategy reaching 98.63% accuracy with zero failed classes on a 3.1M-flow dataset with 2954:1 class imbalance.
|
|
An agentic clinical observability system — a multi-agent command center built on MedGemma, MedASR, and MedSigLIP that autonomously drafts, verifies, and safeguards clinical documentation.
|
A multi-agent Python debugging assistant built on LangChain + Ollama (Llama 3.2) that detects bugs, analyzes code logic, and suggests fixes through a coordinated agent workflow.
|
|
An interactive question-answering bot for a pizza restaurant. Uses LangChain, Ollama-served LLMs, and ChromaDB to answer customer questions grounded in realistic review data.
|
A person re-identification pipeline built in PyTorch with a ResNet-18 backbone for matching identities across camera views.
|
Languages
ML / AI & Agents
Vector Databases
Backend & Infra
