Building AI Systems, Agentic Applications & Python Backends
| Project | Tech Core | Focus & Engineering Highlights | Live / Showcase |
|---|---|---|---|
| ⚡ Peepdesk | Desktop App · Desktop Pet · Electron | Ultra-lightweight Windows desktop companion: a floating animated pet, hydration & chai-break reminders, and an... | Showcase |
| 🤖 TarkaBot SaaS | DeepSeek · DevSecOps · FastAPI | Product code is private. This repository documents the architecture, key decisions and workflow exports. | Live App |
| 🤖 Dispatchai | Agents Sdk · Elevenlabs · Field Service | Voice-first field service dispatch demo — ElevenLabs voice agent, Cloudflare Workers (Agents SDK), n8n... | Showcase |
I'm a Computer Science graduate and AI Systems Engineer focused on building AI applications backed by practical software engineering.
My work centers on Python/FastAPI backends, agentic AI, RAG systems, and multi-tenant SaaS, with additional experience in computer vision and machine learning.
I build systems end-to-end — from API and database architecture to authentication, retrieval pipelines, AI integrations, testing, containerization, and deployment.
- 🤖 Core Focus: Agentic AI, RAG Systems, Python/FastAPI Backend Engineering
- ⚙️ Engineering: PostgreSQL/RLS, REST APIs, asynchronous workflows, Docker, testing & deployment
- 💼 Open to: AI Engineer, AI Software Engineer, Automation Engineer & Backend Engineering opportunities
A private-beta, multi-tenant B2B SaaS that automates restaurant operations. It intercepts WhatsApp orders via a custom Roman Urdu NLP algorithm and manages them through a real-time Kanban kitchen dashboard and POS terminal.
| Component | Technical Implementation Standard |
|---|---|
| 🌐 Live Sandbox | TarkaBot Operating System |
| 🤖 NLP Layer | Custom Roman Urdu parser with structural fuzzy matching matrices |
| 🗄️ Core Database | Multi-tenant PostgreSQL built with strict Row Level Security (RLS) isolation |
| ☁️ DevOps Stack | Containerized workflow via Docker deployed across DigitalOcean, Vercel, & Supabase |
🔗 View System Codebase Showcase
A Computer Science FYP for classifying media as Real, Deepfake, or AI-Generated using custom-trained PyTorch CNN models, with FastAPI inference and a React application layer.
| Component | Technical Implementation Standard |
|---|---|
| 🎯 System Validation | 95.83% Validation Accuracy (XceptionNet-41 on balanced image holdout) · Frame-level temporal voting |
| 🧠 Model Topology | Ensemble execution mode (XceptionNet-41 + EfficientNet-B4 + MobileNetV3) |
| ⚙️ Backend Pipeline | Asynchronous multipart buffer stream routing via FastAPI |
| 🛡️ Memory Defense | Out-Of-Memory (OOM) crash protection decorators & execution timeouts |
🔗 View Model Execution Showcase
A locally deployed Windows AI agent built with LangGraph, capable of web research, file operations, terminal execution, and software management with Human-in-the-Loop approval controls for higher-risk actions.
| Component | Technical Implementation Standard |
|---|---|
| 🤖 Agent Core | Stateful LangGraph orchestrator with persistent SQLite checkpoint memory (.db) |
| 🛡️ Security (HITL) | 3-Tier Human-in-the-Loop approval gate intercepting higher-risk OS terminal calls |
| ⚡ OS Execution | Asynchronous non-blocking subprocess.Popen execution & regex-based command validation guards |
| 🎨 Stark-HUD UI | Glassmorphic desktop interface built with React 19, TypeScript, & Tailwind CSS |
🔗 View System Codebase Showcase
A multi-tenant RAG application designed for source-grounded document Q&A using hybrid retrieval, reranking, and verifiable citations mapping claims to real PDF page numbers and bounding-box coordinates.
| Component | Technical Implementation Standard |
|---|---|
| 🌐 Live Sandbox | Waraq AI Application |
| 🔍 Hybrid Retrieval | Dense (bge-small-en-v1.5) + Sparse (BM25) fused server-side via Reciprocal Rank Fusion (RRF) |
| 🎯 Verifiable Citations | Page-boundary-safe chunking with exact page_number & bounding-box citation metadata |
| 🔐 Multi-Tenant & BYOK | Stateless Bring-Your-Own-Key architecture with Clerk JWT tenant isolation |
🔗 View System Codebase Showcase
A B2B e-invoicing application designed around ZATCA-oriented compliance workflows for businesses working with Saudi Arabian invoicing requirements.
| Layer | System Integration Architecture |
|---|---|
| 🌐 Live Application | ExportShield Enterprise Node |
| 🤖 AI Mapping | SmartMap AI heuristic schema analyzer powered by structured prompt boundaries |
| 🖥️ Node Isolation | Gold Master v1.0.4 Windows runtime binary built with node-locked hardware matching |
| 🔐 Security Matrix | Client-side AES schema encryption, tenant database isolation, & rate-limiting loops |
- 🔍 AuraScout AI — Automated lead-scouting & web intelligence tool: Serper heuristic discovery, Gemini AI schema parsing, BYOK multi-LLM routing (Gemini/Groq/OpenRouter), FastAPI + SQLite.
- 🌌 Project Aura — Forensic reasoning prototype tracking real-time reconstruction of digital artifacts, accelerated via vLLM + ROCm™ 6.0 for AMD Instinct™ MI300X infrastructure (192GB VRAM). Built for the AMD Global Hackathon 2026.
- 🎓 EduMentor Live — Real-time multimodal bilingual tutor leveraging Gemini 2.0 Flash WebSockets and WebRTC bidirectional streaming for low-latency voice and screen context validation. Deployed on Google Cloud Run for the Google Gemini Live API Hackathon.
- 🔒 SentinelFix AI — Multi-agent security remediation prototype built for the GitLab AI Hackathon 2026, combining Claude 3.5 Haiku scanning loops with automated Claude 3.5 Sonnet codebase remediation agents.
- 💬 Customer Support Chatbot — NLP customer-support automation: GPT-2 empathetic reply generation with VADER/TextBlob sentiment detection and TF-IDF intent classification. Deployed live on Streamlit Cloud.
- 🩺 MediVision AI — Disease Predictor — End-to-end healthcare ML system (scikit-learn): data preprocessing → multi-model training → user-facing Streamlit application.
- 🛡️ Fraud Detection ML — Final internship project: ensemble anomaly detection (Isolation Forest + K-Means) evaluated on 17.5K job applications (94.2% precision, 91.9% F1 on validation split) with an interactive Streamlit monitoring dashboard.
- AI & Agentic Systems: LangGraph, LangChain, PyTorch, RAG Pipelines, Vector Indices (Qdrant, Pinecone, PGVector, ChromaDB), Computer Vision (OpenCV, MediaPipe).
- Backend Engineering: Python, FastAPI, Node.js, PostgreSQL (RLS Design), REST APIs, Asynchronous Workflows, Connection Pooling, Redis Cache Layers.
- Frontend Core: React 19 / 18, Next.js 14, TypeScript, Tailwind CSS, WebRTC, WebSocket Bidirectional Streaming.
- DevOps & Cloud: Docker Containerization, GitHub Actions CI/CD, Google Cloud Run, DigitalOcean, Vercel, Netlify.
