Computer Engineer · Islamabad

Maria Lateef

I build machine-learning systems that reach off the screen.

From neural networks and LLM applications to the robots, sensors, and devices that run them. Trained models, soldered boards, and the Python that ties the two together.

About

I am a Computer Engineering graduate who takes projects the whole way, from training and validating a model to deploying it on real hardware. My work spans deep learning and generative AI, embedded systems, IoT, and robotics, and I like the problems that sit where software meets the physical world.

Degree
BS Computer Engineering
University
COMSATS Islamabad
CGPA
3.37
Recognition
3rd · Career Expo
Selected Work

Projects that go from model to machine.

Flagship · Deep Learning + Hardware

Handwriting Emulator System

Final year project · 3rd place, COMSATS Career Expo

An end-to-end system that turns typed or spoken text into realistic human handwriting, physically drawn on paper by a custom-built robotic pen plotter. Fully offline, with six selectable styles. I trained a 1.89M-parameter LSTM-MDN model with Graves-style soft attention on the IAM database, diagnosed and fixed an attention-collapse defect with a kappa-velocity penalty, built a dual-mode OCR pipeline (Tesseract and Microsoft TrOCR) with Whisper voice input, and designed the two-axis plotter hardware and its stroke-to-G-code pipeline.

PyTorchLSTM-MDNAttention TrOCRWhisperArduino / GRBLStreamlit
Generative AI · RAG

Pakistan Case-Law Assistant

A prototype AI research assistant that retrieves relevant Supreme Court of Pakistan judgments for a typed or spoken question, summarises each with an LLM, and labels which side it supports. Built on a RAG pipeline with sentence-transformer embeddings, a Chroma vector database, and Google Gemini for summaries, side-labelling, and voice transcription.

RAGLLMsGemini Vector DBStreamlit
Machine Learning · Security

IoT Attack Detection

A machine-learning system that detects and classifies cyberattacks on IoT devices such as cameras and sensor nodes. I trained, benchmarked, and compared Random Forest, SVM, and Decision Tree classifiers on the IoT-DH dataset, validating accuracy across attack categories. Co-authored a research paper (under review).

scikit-learnPandasClassification
Embedded · IoT · Robotics

Hardware & IoT

A real-time IoT water-level monitor on ESP32 with remote MQTT alerts and a scalable topic architecture, plus two Arduino competition robots: a sumo bot with IR edge detection and a PID-tuned line-following bot. Firmware, sensors, and control logic, hands on the hardware.

ESP32MQTTArduino PIDC++
Cloud

Cloud & Infrastructure

Hands-on cloud work through the DecodeLabs internship: provisioning and securing a Linux web server on AWS EC2 from scratch, hosting static sites on S3 and Azure Blob Storage, and configuring everything over SSH from the command line.

AWS EC2 / S3AzureLinuxNginx
Toolkit

What I work with.

Machine Learning & AI

Python · PyTorch · deep learning · LSTM / RNN · mixture density networks · attention · scikit-learn · OpenCV · NumPy · Pandas

Generative AI

LLMs · Retrieval-Augmented Generation · Google Gemini · embeddings · vector databases · Hugging Face · prompt design

Embedded · IoT · Robotics

Arduino · ESP32 · MQTT · GRBL · PID control · ultrasonic & IR sensors · stepper & servo motors · G-code

Cloud · Data · Languages

AWS (EC2, S3) · Azure · MySQL · Linux · Git · Streamlit · C · C++ · SQL

Contact

Open to roles in AI, ML, software, and embedded / hardware. Let's talk.