Muhammad Ahsaan Ullah
AI Systems Engineer | Python/FastAPI · Agentic AI · RAG
BS Computer Science Graduate, University of Management and Technology, Lahore
Building AI agents, workflow automation, and Python/FastAPI backends designed for async and concurrent workloads.
Engineering Philosophy
I build end-to-end AI applications with practical software-engineering foundations: typed APIs, data validation, authentication and tenant isolation, asynchronous workflows, retrieval/evaluation pipelines, testing, containerization, and deployment.
Every system is engineered with clear architectural boundaries: tenant-isolated PostgreSQL architectures using Row Level Security (RLS), asynchronous FastAPI microservices, deterministic agent execution loops, and automated test coverage.
Working Tools. In Production.
Tools I use daily to build production systems: a battle-tested working stack, not a checklist.
Backend & Cloud
Multi-cloud & Backends
AI & Vector Engines
Agents, RAG & Vectors
Auth & Integrations
Real-time & Security
Web & Hosting
Client platforms
Portfolio
Featured Projects
Live systems engineered with production discipline, automated testing, and strict architecture.
A multi-tenant RAG application designed for source-grounded document Q&A using hybrid retrieval, reranking, verifiable page-level citations, and 55 automated unit/regression tests.
Locally deployed Windows AI agent built with LangGraph for web research, file operations, terminal execution, and workflows, featuring strict Human-in-the-Loop approval gates and 68 automated tests.
Private-beta multi-tenant restaurant SaaS turning WhatsApp orders into structured kitchen/POS operations using FastAPI, React, PostgreSQL RLS tenant isolation, Roman Urdu parsing, and 32 tests.
AI-assisted candidate screening workspace for recruiters, designed to evaluate resumes with evidence-backed scoring, structured screening workflows, and safeguards against prompt-injection-style manipulation.
DeepFakeShield
Computer Vision FYPComputer Vision Final Year Project for classifying media as Real, Deepfake, or AI-Generated using PyTorch CNNs, FastAPI inference, and frame analysis (95.83% XceptionNet-41 accuracy on a balanced image holdout).
B2B e-invoicing application designed around ZATCA-oriented compliance workflows at exportshieldpro.online, with ECDSA PKI signing, XML/UBL validation, and sandbox integration.
Timeline
Experience
Architected and deployed AI systems, agentic automation pipelines, and asynchronous backend platforms for contract clients and independent SaaS deployments:
- Waraq AI (waraqai.netlify.app): Deployed multi-tenant hybrid RAG system with verifiable page-level citations and 55 automated tests.
- Agent Friday: Locally deployed Windows AI agent using LangGraph state machines, terminal execution, and human-in-the-loop safety gates (68 tests).
- TarkaBot (tarkabot.online): Private-beta multi-tenant Restaurant SaaS automating WhatsApp orders, kitchen KDS, and delivery dispatch with PostgreSQL RLS.
- ScreenOS: AI-assisted candidate screening workspace for recruiters, designed to evaluate resumes with evidence-backed scoring and prompt-injection safeguards.
- DeepFakeShield: Computer Vision FYP for synthetic media classification achieving 95.83% XceptionNet-41 accuracy on a balanced image holdout.
- ExportShield Pro (exportshieldpro.online): ZATCA-oriented e-invoicing middleware with cryptographic ECDSA PKI signing and XML/UBL validation tested in the ZATCA sandbox.
Built Power BI analytical dashboards and complex SQL queries; automated recurring reporting pipelines and data extraction workflows.
Built ML and NLP models in Python; created interactive Streamlit dashboards to validate model results and streamline evaluation workflows.
Carried out data cleaning, exploratory data analysis (EDA), and feature engineering pipelines for predictive modeling.
Background
Education & Certifications
BS Computer Science
University of Management and Technology (UMT), Lahore