Hi, I'm Malik Mohammad Ammar

Full Stack AI Engineer

I build AI systems that reason, see, and move: agentic search, real-time vision, and robotics.

About

I'm a full-stack AI engineer who ships end to end, from model to interface. I've built an agentic search engine on ReAct reasoning and multi-source RAG, deployed object detection and Re-ID tracking on NVIDIA Jetson, and worked across ROS 2, PX4, and embedded hardware to connect software to the physical world.

I also build the product around the model: backend infrastructure, databases, and modern web interfaces. Whether it's an AI search engine, a vision system, or a consumer platform, I like turning a hard technical idea into a tool people can use.

I do my best work where research meets production: reasoning systems, real-time perception, and robots that have to work outside a notebook.

Education

National University of Sciences and Technology (NUST)

2020 - 2024

Bachelor of Engineering, Electrical Engineering

Certifications

Oracle Cloud Infrastructure 2025 AI Foundations Associate

Oracle · Oct 2025

Python Essentials 1

Cisco · Sep 2024

Machine Learning for Data Science Projects

IBM · Aug 2024

AI Skills Challenge

Microsoft Learn · Aug 2023

Experience

Research Engineer

School of Interdisciplinary Engineering & Sciences (SINES)

May 2025 - Aug 2025

Deployed object detection and ID-based Re-ID tracking on NVIDIA Jetson, with a low-latency GStreamer and OpenCV pipeline for real-time inference.
Fine-tuned YOLO11n on a custom aerial and ground dataset, reaching 51% mAP for people and 66% for cars, then converted it to TensorRT for Jetson deployment.
Designed SITL and HITL simulations with AirSim, PX4, and ROS 2 to validate autonomous quadrotor navigation before flight testing.

Computer Vision

NVIDIA Jetson

GStreamer

OpenCV

TensorRT

PyTorch

Intern

Public Sector Research Lab

Jul 2023 - Sep 2023

Built a hybrid skin-care recommender on the Sephora dataset, combining user-feature filtering with TF-IDF and cosine similarity.
Evaluated Naive Bayes, logistic regression, and SGD classifiers, reaching 98.38% accuracy with Precision@10 and NDCG@10 both at 1.00.

Python

Scikit-Learn

NLP

TF-IDF

Tkinter

Skills

Programming Languages

Python

C/C++

JavaScript

TypeScript

Verilog

Machine Learning / AI

PyTorch

TensorFlow

Scikit-learn

NumPy

Pandas

SciPy

OpenCV

Hugging Face

LangChain

Web Development

React

Next.js

Node.js

Tailwind CSS

shadcn/ui

Mobile Development

React Native

Expo

Backend & Cloud Platforms

Supabase

Render

Streamlit

Robotics & Simulation

ROS 2

PX4

AirSim

Edge AI / Deployment

NVIDIA Jetson

TensorRT

DeepStream

Developer Tools

Git

GitHub

Docker

Projects

Check out my latest work

OmniFinder AI

Agentic search engine that reasons with ReAct and answers from multi-source RAG. Routes queries across Wikipedia, ArXiv, and the web, with a Streamlit interface on top.

Python

LangChain

Streamlit

LLM

RAG

ArXiv

DuckDuckGo

Salert

Deal discovery platform for Pakistan that aggregates sales and discounts across clothing, food, and consumer goods, with geo-targeted deals, brand following, and voting.

Next.js

React

TypeScript

Tailwind CSS

Supabase

Vercel

Autonomous Quadrotor Object Tracking System

Real-time object detection and Re-ID tracking on NVIDIA Jetson with TensorRT and ONNX. Validated the autonomous control loop through SITL and HITL simulation.

Python

OpenCV

PyTorch

YOLO

ROS

PID Control

TensorRT

Smart Cane for Visually Impaired

Assistive cane that detects obstacles in real time on Raspberry Pi 5, combining YOLO object detection with ultrasonic sensors and audio alerts for safer navigation.

Python

OpenCV

PyTorch

YOLO

Raspberry Pi

Ultrasonic Sensors

Product Recommendation System

Hybrid recommender ranking products by TF-IDF and cosine similarity. 98.38% accuracy, with Precision@10 and NDCG@10 at 1.00.

Python

NLP

Scikit-Learn

Cosine Similarity

Tkinter

Heart Disease Severity Classification

Classified heart disease severity with 94.4% accuracy using an SVM. EDA and PCA flagged triglyceride level as the strongest predictor.

Python

SVM

MLP

Random Forest

Scikit-Learn

PCA

Get in Touch

Open to remote AI engineering roles and research collaborations. Email is the fastest way to reach me.

malikmammar17@gmail.com Résumé