Every project here answers one question: what problem does this system reason about? From 400,000-record national databases to enterprise AI pipelines — built to perform.
11Public Systems
1AI / ML
3Data Systems
7Full‑Stack
Live from GitHubMasum-Khan003·Auto updates every day
A policy enforcement runtime that converts enterprise authorization decisions into retrieval-time security constraints and independently verifies the evidence before it reaches an AI system.
Problem
A policy enforcement runtime that converts enterprise authorization decisions into retrieval-time security constraints and independently verifies the evidence before it reaches an AI system.
System
Built with Python as the core implementation language, structured for clarity and reuse.
Behavioral guardrails for tool-calling AI agents - a framework-agnostic interception layer that watches an agent's full action trajectory, not just a single prompt, to catch the moment a legitimate-looking tool call is actually driven by an instruction the user never gave.
Problem
Behavioral guardrails for tool-calling AI agents - a framework-agnostic interception layer that watches an agent's full action trajectory, not just a single prompt, to catch the moment a legitimate-looking tool call is actually driven by an instruction the user never gave.
System
Built with Python as the core implementation language, structured for clarity and reuse.
E-CIP is a solo-built, production-grade ML platform that demonstrates end-to-end machine learning engineering, not just model notebooks, but the full lifecycle: data validation, training, calibration, explainability, serving, observability, and a live operations dashboard.
Problem
E-CIP is a solo-built, production-grade ML platform that demonstrates end-to-end machine learning engineering, not just model notebooks, but the full lifecycle: data validation, training, calibration, explainability, serving, observability, and a live operations dashboard.
System
Built with Python as the core implementation language, structured for clarity and reuse.
A production-grade ML platform that scores financial transactions for fraud in real-time at <150ms p99 with full explainability, drift monitoring, and a 4-panel Streamlit dashboard.
Problem
A production-grade ML platform that scores financial transactions for fraud in real-time at <150ms p99 with full explainability, drift monitoring, and a 4-panel Streamlit dashboard.
System
Built with Python as the core implementation language, structured for clarity and reuse.
A production-ready Fake News Detection system built using Deep Learning (BiLSTM) and deployed via a scalable API architecture. This project transforms an experimental Jupyter Notebook model into a real-world inference service with API integration and frontend compatibility.
Problem
A production-ready Fake News Detection system built using Deep Learning (BiLSTM) and deployed via a scalable API architecture. This project transforms an experimental Jupyter Notebook model into a real-world inference service with API integration and frontend compatibility.
System
Built with Jupyter Notebook as the core implementation language, structured for clarity and reuse.
