I'm a GenAI Engineer focused on building practical AI systems, from LLM-powered applications and multi-agent workflows to intelligent automation.
My experience spans LLMs, RAG, agent orchestration, LangChain, LangGraph, and MCP. I'm expanding my hands-on expertise in classical machine learning, model training, feature engineering, and evaluation, with a focus on solving real-world problems through data-driven approaches.
I enjoy taking ideas from concept to implementation, experimenting with new techniques, and building systems that go beyond demos to deliver practical value.
Always building, always learning, and exploring the intersection of GenAI, ML, and data science.
Dec 2025 – Jul 2026
My first end-to-end ML project, benchmarking Logistic Regression, Decision Trees, Random Forest, Gradient Boosting, and XGBoost for binary network intrusion detection on the UNSW-NB15 dataset. Explored feature preprocessing, hyperparameter tuning, precision/recall/F1, and feature importance, achieving a best test accuracy of 96.34% with XGBoost.
Personal AI Assistant Framework. Architected a multi-user AI assistant backend supporting streaming text and real-time voice interaction using FastAPI, LangGraph, and WebSockets. Designed a three-tier persistent memory system (Profile / Episodic / Semantic) backed by PostgreSQL and ChromaDB with composite retrieval scoring using semantic similarity, importance weighting, and recency decay. Implemented dual retrieval runtimes (semantic vector retrieval and lightweight lexical retrieval) switchable at runtime, built scalable agent orchestration pipelines with contextual memory injection, and containerized backend workloads on ACI integrating Groq STT, finetuned GPTSoVits TTS, and local embedding runtimes.
A modular persistent memory framework for LLM agents. Provides a four-tier memory architecture backed by PostgreSQL and ChromaDB, exposed as a FastAPI REST service.
An autonomous Web Application Penetration Testing framework built with LangGraph, featuring a dual-planner system with adaptive plan-execute-replan loops across recon, enumeration, exploitation, and reporting. Uses MCP-backed agents for Nmap, HTTP inspection, and XSS validation, with parallel task execution and shell-injection protection. Validated against TryHackMe lab environments.
A basic Endpoint Detection and Response project focused on device registration and management. A Python agent collects system information (hostname, OS, serial, CPU, RAM, GPU, network interfaces) and registers devices with a FastAPI server storing to Supabase PostgreSQL. Includes a React admin dashboard with JWT authentication, logout blocklisting, and protected device endpoints.
An educational web application showing how XSS and SQL Injection work. Built to teach students and beginners how insecure input handling can be exploited, with clear examples of payloads and insecure patterns.
A hand-gesture detection system that converts real-time webcam keypoints into keyboard shortcuts. Built for productivity and automation, featuring gesture recognition, shortcut mapping, and smooth real-time interaction.