I build agentic AI systems, and I test them like production software.
MSc Artificial Intelligence at BTU Cottbus, Germany. Before that, three years of professional QA, test automation, and project delivery in telecom and banking. That combination is the point: most people arriving at agentic AI come from ML or web — I came from being professionally accountable for whether software is correct, applied to systems whose output is non-deterministic.
📍 Cottbus, Germany · open to remote and relocation across Europe · valid German work authorization 🔎 Looking for junior–mid roles in AI / agentic systems / automation or QA & test automation
🛒 NoSiappa.pk — peer-to-peer marketplace with escrow-protected payments
A production marketplace for used smartphones, built solo, end to end. Integer-paisa double-entry ledger, escrow state machine, maker/checker separation of duties on fund recognition, and a device-verification layer built around a national registry that has no public API.
Next.js 15 tRPC Prisma PostgreSQL TypeScript Vercel Sentry
| Built in | 52 days · 396 commits · 42 active days |
| Scale | 932 files · ~79,700 lines of TS/TSX · 53 data models · 37 migrations |
| Tests | 2,339 across 204 files |
| Agent-assisted | 377 of 396 commits — I reviewed and approved every one. No autonomous merges. |
🛡️ LLMFence — guardrail API for LLM input and output
Give it text, get back a risk score, a status, the flags that fired, and a redacted version. Metered credit billing, tiered rate limiting, non-blocking telemetry. Shipped with an 8-case adversarial QA suite — including a prompt-injection payload that instructs the analyzer itself to return a clean score, which it correctly rejects at risk 100, and a false-positive stress case that keeps the detector honest.
Python FastAPI Next.js React PostgreSQL Supabase Railway Docker
📊 Read the engineering case studies → — architecture, the bugs that got caught before production, and what agent-assisted development actually cost to run.
| Systems Limited · Consultant, QA | Software Integration Testing for Etisalat UAE (e&). Selenium→Playwright framework migration, API and data validation across 6–8 interconnected systems in the SIM provisioning chain. ~30% reduction in manual regression effort. |
| Faysal Bank · Regulatory Projects → Cards & Digital Ops | UAT ownership on VISA/MasterCard gateways at 90% first-pass. Automated a portal signup to create 300+ user IDs in one night during a live remittance outage, then documented the framework and trained the permanent team. |
| MSc Artificial Intelligence, BTU Cottbus | ML, NLP with PyTorch, applied AI. Two courses and thesis remaining. |
Certified: ISTQB® Foundation Level (CTFL 4.0) — 92.5%
Languages Python · TypeScript · JavaScript · SQL AI / agents Model Context Protocol · agent skills & orchestration · persistent agent memory · prompt engineering · NLP with PyTorch Backend FastAPI · tRPC · Prisma · PostgreSQL · Supabase Frontend Next.js · React · Tailwind Testing Playwright · Selenium · PyTest · Postman · TDD/BDD Infra Docker · Vercel · Railway · Jenkins · GitHub Actions · Sentry
My two flagship projects are in private repositories — they carry production business logic, identity-verification flows, and money-path code, and publishing them would be a security problem, not a portfolio win.
So instead: the products are live and clickable above, and the case studies repo documents the architecture, the decisions, and the measurements in public. Source access for a specific hiring process is something I'm happy to arrange directly.
The older public repos on this profile are student-era work from 2017–2024 and aren't representative of current output.