Profile sync: 2026-10-20
Building machine learning systems and backends where every number is computed, tested and true
I'm an ML and backend engineer and a Computer Science graduate of the University of Central Punjab. I build machine learning systems and the backends that carry them into production, from data pipelines and trained models to APIs, databases and the product a person actually uses.
Most of my work starts from one observation: a wrong number looks exactly like a right one on the page. So I let code compute what can be computed, verify what I build before I trust it, and use a language model only where it earns its place. That is how I ended up building an analytics assistant that never invents a figure, and a resume tool that refuses to invent a skill you don't have.
Right now, I'm leading the AI and machine learning track at Digitech Offerings, where I designed and delivered an eight-week AI/ML programme for forty students, while working as a freelance software engineer building web platforms, backends and AI features for clients. The rest of my time goes into products that put correctness first: InsightFlow AI analyses business data with a deterministic engine, RoleFit rewrites resumes without fabricating a single claim, and JARVIS runs a full voice assistant entirely on one laptop. In between, I publish reproducible studies on how machine learning evaluation quietly misleads, and I'm open to full-time roles, freelance work and research collaborations.
|
Core Competencies Languages |
AI / ML Stack Data & Visualization Web & Frontend |
Cloud & Infrastructure Databases & ORMs Dev Tools |
graph LR
A[Real Problem] -->|Understand the data| B[Deterministic Core]
B -->|Model only where it earns it| C[ML / LLM Layer]
C -->|Check every output| D[Guardrails]
D -->|Validate on real data| E[Evaluation]
E -->|Docker + CI| F[Deploy: Vercel / Azure]
F -->|Measure and iterate| B
style A fill:#1a1b27,color:#6C63FF,stroke:#6C63FF
style F fill:#1a1b27,color:#6C63FF,stroke:#6C63FF
|
AI that computes its facts and refuses to invent them InsightFlow AI · RoleFit · JARVIS |
Models judged by honest evaluation, not a lucky split CKD Prediction · Leakage Demo · Grad-CAM Study |
Secure, tested systems from schema to shipped product Locora · Momentum · Portfolio |
|
|
|
|
|
|
|
|
|
Every engineer has a unique signal. Here's mine.
# ── SIGNAL PROFILE ──────────────────────────────────────────────
handle : "Zain Ul Abideen"
origin : "Pakistan 🇵🇰"
signal : "ML & Backend Engineering"
transmitting:
- "AI products where code computes the facts and models phrase them"
- "Backends and APIs built for production, not just notebooks"
- "Local-first AI that keeps personal data on the device"
- "Honest evaluation, because a lucky split is not a result"
frequency : "Always learning. Always shipping."
bandwidth : "Deep focus + open collaboration"
wavelength : "Where research meets real-world impact"
# ────────────────────────────────────────────────────────────────"The best engineers are those who collaborate, communicate, and compound their knowledge."






