I completed a Computing Science degree at the University of Alberta in June 2026. I make backend services, data pipelines and systems that use a language model.
My aim is to make these systems reliable. Each system has tests and checks that run after each change. A log shows the reason for each decision. Each project below has a demo that you can open. Each project also shows all its results, and this includes the bad results.
Portfolio: https://samad-zeeshan.github.io/abdul-samad-zeeshan-portfolio/
- Warden examines changes to live software. It accepts safe changes and stops dangerous changes. An AI assistant that gets false instructions cannot make Warden accept a bad change. Demo · Repo
- Proving tests a new version of an AI agent on thousands of test customers. Then it tells you to release the agent or to stop it. It does this before the agent speaks to a real customer. Demo · Repo
- Tally is a small bank. It does not lose a cent, and it does not count a cent two times. I checked it with a formal model. I also checked it with network failures and database failures. Demo · Repo
- Parley answers phone calls and books visits to properties. It speaks English and Gulf Arabic. It gives only facts from the listings database. Demo · Repo
- Tarn finds a known attacker in one billion real logins. Then an AI analyst puts the alerts in order for a human analyst. Demo · Repo
- Triage is a small offline model. It sorts support emails into JSON. Before training, it gave the correct urgency to 38% of emails. After training, it gave the correct urgency to 87% of 1,200 new emails. Demo · Repo
- Docket reads photos of receipts and changes them into checked data. With an error limit of 1%, it tells you which receipts do not need a human check. Demo · Repo
- Bourse is a small stock exchange. It does not lose an order. Strict tests show if the profit of its trading robot comes only from luck. Demo · Repo
My older projects are in my repositories. These projects include Fulcrum, Touchstone, Kvasir and pathfinder.