AI / ML Engineer
I train the model, wrap it in a pipeline that retrains and redeploys it, and automate everything either side of that.
Work
Pulled live from GitHub, so this list stays current.
Full-stack
MERN+ Docker deploy
Hospital management + patient portal, dockerised
An enterprise-shaped hospital management system and patient portal on the MERN stack, containerised with Docker. Role-separated: staff manage records and scheduling, patients see only their own care.
Machine learning
A set of working GenAI tools, not demos
A growing collection of Streamlit apps wrapping generative models into things a non-technical person can actually use — each one a small tool that does a single job end to end.
Machine learning
Predicts machine failure before it happens
A predictive-maintenance system for industrial equipment: deep-learning models score sensor streams for time-to-failure, and the whole thing serves behind an API with live monitoring. The full lifecycle in one repo — training, serving, and watching it in production.
Systems
Manualsyscalls + memory
A file system, its syscalls, and a shell — in C
An educational operating-systems build: a virtual file system in C with manually implemented system calls, its own memory management, and a working shell on top. This is the project that taught me what the abstractions above it are actually hiding.
Security
6crypto primitives implemented
Cryptographically tamper-proof banking ledger
A secure banking platform where every transaction is signed and chained. Implements RSA, AES-256, Diffie–Hellman key exchange, digital signatures, PKI certificates, and hybrid encryption on top of a proof-of-work blockchain, so the transaction history can't be quietly rewritten.
Machine learning
4models, zero libraries
Four classifiers built in NumPy — no scikit-learn
A diabetes risk predictor on the CDC Health Indicators dataset where every model is implemented by hand: decision tree, naive Bayes, KNN, and a feed-forward network, all in NumPy. Written this way on purpose — the point was to own the gradient maths and the splitting criteria rather than call .fit() on them.
Machine learning
End-to-endscrape → model → GUI
Scrapes Zameen.com, prices Islamabad property
An end-to-end pipeline that scrapes live property listings from Zameen.com, cleans the very messy real-world result, and predicts house prices in Islamabad. Wrapped in a CustomTkinter desktop app so someone non-technical can actually use it.
Machine learning
10scenarios covered
Learns which search algorithm to use
Rather than hard-coding one pathfinding strategy, this picks the right search algorithm for the situation — a model trained across ten delivery scenarios recommends the approach that will actually perform best on each.
Machine learning
RNNs, LSTMs and GRUs from the maths up
LSTM and GRU worked through from the mathematical foundations to TensorFlow implementations, then applied to real NLP tasks. Written the way I learn things: derive it first, then build it, then use it.
Machine learning
Image models, from classifiers to detection
A working set of computer-vision notebooks — building up image classification and detection models and the preprocessing pipelines that feed them.
Machine learning
Applied ML work during my internship
Notebooks and deliverables from my AI/ML internship at ITSimplera Institute: applied model building against real assignment briefs.
Practice
Algorithm practice in C++, most days
An ongoing log of data-structures and algorithms practice in C++ — the maintenance work that keeps problem-solving sharp.
Security
Information-security-first transaction ledger
A ledger system designed around information security principles — integrity, auditability, and controlled access to financial records. The groundwork that AegisChain grew out of.
Practice
Scaffold for the delivery routing work
The Python skeleton I built out the Islamabad delivery routing logic on top of.
Systems
O(E log V)route solving
Dijkstra over a live delivery graph
A C++ management system built directly on the data structures that make it work: graphs for the delivery network with Dijkstra for shortest-path routing, AVL and BST trees for indexed lookups, and linked lists for order queues.
Machine learning
Working through the framework, layer by layer
My PyTorch notebook collection — tensors, autograd, custom nn.Modules, training loops written out rather than abstracted away. This is where the from-scratch maths meets the framework that does it faster.
Machine learning
Core algorithms, worked by hand
Notebook work covering the classical ML ground: regression, classification, clustering, and the evaluation habits that go with them.
Systems
Losslessverified round-trip
Lossless compression in 32-bit x86 assembly
A run-length encoding compression engine written entirely in MASM x86 assembly, with lossless round-trip verification and per-session analytics on compression ratio. The lowest level anything here goes.
Full-stack
5-dayforecast horizon
Real-time weather, React + FastAPI
A full-stack weather dashboard: FastAPI backend serving current conditions and a five-day forecast for any city, React frontend rendering it. Clean split between the data layer and the interface.
Practice
Keras and TF hands-on notebooks
Practical TensorFlow and Keras work — model building, training, and evaluation, kept as a running reference alongside the PyTorch notebooks.
Journey
Each epoch is a step down the curve — the loss being the gap between what I could build and what I wanted to.
Toolkit
About
I'm Hasnat, an AI/ML engineer based in Islamabad. I build the whole loop: the scraper that pulls data nobody packaged, the model trained on it, the pipeline that retrains and redeploys it when it drifts, and the API or GUI a non-technical user can actually run.
I came at this from underneath — x86 assembly, C, manual system calls, cryptography, data structures by hand — then NumPy classifiers, then PyTorch. That scaffolding is why I can tell a bad gradient from a container that won't start, and why the automation I write doesn't just call .fit() and hope.
Currently interning in AI/ML and looking for the next place to build production pipelines that run themselves.
ITSimplera Institute
Selected for an applied AI/ML internship — building and evaluating models against real project briefs.
Self-directed
Implemented four classification algorithms end-to-end in NumPy, including hand-derived backpropagation, rather than using a library.
Self-directed
Shipped a banking platform implementing six cryptographic primitives on a proof-of-work chain for tamper-evident transaction history.
Connect
Open to AI/ML, MLOps and automation roles. The fastest way to reach me is email — I read everything.