Software Engineer
Building calm, durable systems that work under pressure.
Profile
I'm Abdullah, a software engineer. I'm more interested in what a system does on day 90 than on day 1, the boring-but-load-bearing layer where retries, idempotency, and failure modes decide whether software stays useful or just demos well. Most of my work has drifted from pure backend toward end-to-end development and LLM/agentic systems, and the interesting problems live in the same place: making AI behave in production. Currently building Vendor Master Intelligence at PieCyfer.
Core tools
Selected work
An AI system that finds what's hiding in messy procurement data — reading 30K+ contracts per client, reconciling them against ERP records, and flagging duplicates, pricing anomalies, and compliance risks. Surfaced over $160K in recoverable savings on the first full run across $2.1B in vendor spend.
A learning platform that turns your own course materials into adaptive quizzes, diagrams, and flashcards on demand. RAG over your documents, multi-model LLM routing, and tool calls that pick the right artifact for whatever you're studying.
A full-stack voice agent platform — the kind a real business could pick up tomorrow. Real-time STT/LLM/TTS over LiveKit with multi-turn dialogue, intent classification, call routing, booking flows, and conversation analytics. FastAPI backend, React/TypeScript frontend.
A regional data ingestion pipeline replacing a manual weekly process, someone checking a dozen+ listing sites by hand, with a self-running system that builds the canonical data layer for downstream product analytics. Built on Scrapy for static targets and Playwright (with managed proxies and stealth) for JS-heavy ones, running at 10K+ pages/day at 85%+ extraction accuracy. State-by-state rollout: the initial Texas run pulled 500K+ records into a unified schema; Louisiana is next. The interesting architectural piece is a depth-aware RabbitMQ queue that backpressures crawl rate against MySQL write capacity, so the database never becomes the bottleneck no matter how fast the spiders finish.
18 min read
What three months of building a contract-intelligence pipeline taught me about production AI systems.
1 min read
First post on the log.
Trajectory
PieCyfer
Labmise
Broadstone Technologies
Softturf
Foundation
University of Management and Technology, Lahore