I finished my Computing Science degree at the University of Alberta in June 2026. Most of what I build is backend services, data pipelines, and systems with an LLM somewhere inside.
What I enjoy most is making that stuff hold up. That means real tests, evals that run on every push, retries that never double-count anything, and a record you can follow to see why a decision was made. Every project below has a demo you can click and reports its numbers as they came out, including the bad ones.
Portfolio: https://samad-zeeshan.github.io/abdul-samad-zeeshan-portfolio/
- Warden: a guard for changes to live software. It approves the boring ones, blocks the dangerous ones, and cannot be talked round by a tricked AI assistant. Demo · Repo
- Proving: tests a new version of an AI agent on thousands of made-up customers before it meets a real one, then says ship or hold. Demo · Repo
- Tally: a small bank on a double-entry ledger that never loses or doubles a cent, checked with TLA+ and fault runs on Kubernetes. Demo · Repo
- Parley: a phone agent that books property viewings in English and Gulf Arabic and only states facts the listings database returned. Demo · Repo
- Tarn: a login graph scored against a billion real records with a known attack, plus an AI analyst that sorts the alerts. Demo · Repo
- Triage: a 600M-parameter model distilled to sort support emails into JSON offline. Priority accuracy went from 38.3% to 87.3% on 1,200 held-out tickets. Demo coming · Repo
- Docket: reads receipt photos into schema-checked data and says which ones can skip a person at a 1% error budget. Demo · Repo
- Bourse: an order matching engine and market maker. In progress, built by hand.
Older work, including Fulcrum, Touchstone, Kvasir and pathfinder, is in my repositories.