Good at turning research into working systems

AI/ML • LLM & RAG • AR/VR • Research

Muhammad Talha

Founder and inventor of AR-CAD, and an AI/ML engineer who builds intelligent products end to end: LLM and RAG systems serving 22 clients, computer vision for manufacturing, and EEG signal processing with signal reconstruction.

Portfolio CV View Projects

Contact me →

About Me

Build AI products end to end

My strength is carrying an idea from research through model building, evaluation, and deployed software, then proving it works: a 20-participant controlled study, a peer-reviewed publication, 22 paying clients, and CAD $25K in commercialization funding.

What I do

I am a computer scientist and machine learning engineer who builds AI products end to end, from data pipelines and model training to evaluation and deployed software. I am the founder and inventor of AR-CAD, an augmented-reality design system validated with a 20-participant controlled study, and a co-founder at WarpShift, where our LLM and RAG systems automate document-heavy legal and real-estate work for 22 paying clients.

I apply machine learning across domains: NLP pipelines that auto-coded 1,800+ think-aloud utterances, computer vision for manufacturing partners (YOLOv8; 97% held-out spectral classification), EEG signal processing with clean signal reconstruction, deep-learning time-series forecasting, and diffusion models for generative content.

What ties it together: I do not stop at the model. I run the user study, build the evaluation harness, ship the software, and publish the results.

Resume Portfolio CV Contact

Core Skills

AI/ML + LLMs

Python PyTorch Scikit-learn Hugging Face Transformers LLM APIs (Anthropic / Gemini / Groq) RAG Prompt Engineering ChromaDB (Vector DB) Text-to-SQL Sentence-BERT Whisper spaCy YOLOv8 OpenCV Diffusion Models Model Evaluation Pandas / NumPy

Data + Signal Processing

SQL Time-Series Forecasting Blocked Cross-Validation EEG Signal Processing MNE-Python ICA Artifact Removal Streamlit

XR + Software + Research

Unity C# Meta Quest 3 Meta XR SDK SteamVR 3D Interaction C/C++ JavaScript / TypeScript React Node.js REST APIs Git / GitHub User Studies Usability Testing (SUS / NASA-TLX) Figma

Education

My academic background combines engineering, computer science, machine learning, human-computer interaction, computer graphics, Neuroscience, and research-driven product development.

Graduate Degree

University of Alberta

Jan 2024 – Apr 2026

Master of Science in Engineering Management (Thesis-Based), Applied AI-focused

Edmonton, Canada

Thesis: How Interaction Modality Shapes Design Cognition: Evaluating Augmented Reality CAD Through Behavioral, Cognitive-Process, and Neurophysiological Evidence.

Supervisors: Dr. Ahmed Jawad Qureshi and Dr. Sehrish Javed

Thesis research focused on AR-CAD, immersive interaction, comparative CAD evaluation, Bloom-coded think-aloud analysis using NLP, and EEG-based cognitive-load assessment for design tasks.

GPA: 3.8/4.0 AR/VR AI/ML HCI EEG

Relevant Courses

Applied ML & AI Machine Learning for Procedural Content Generation Wearables & IoT and Machine Learning Intellectual Property & Commercialization Ergonomics & Work Design

Teaching

TA — Applied Machine Learning & AI (ENGM 680) TA — Statistics & Data Management for Engineers (MEC E 590) Guest Lectures — MEC E 265 (interaction modality & AI in design) Marking — Mechanical Measurements & Numerical Methods

Undergraduate Degree

Lahore University of Management Sciences

Sep 2019 – Jun 2023

Bachelor of Science in Computer Science

Lahore, Pakistan

Built a strong computing foundation across software engineering, algorithms, databases, machine learning, artificial intelligence, HCI, computer graphics, systems, and networking.

Computer Science HCI Graphics ML/AI

Relevant Courses

Object-Oriented Programming Data Structures Algorithms Operating Systems Software Engineering Machine Learning Artificial Intelligence Human-Computer Interaction Information Communication Technology for Development Topic in Human-Computer Interaction Computer Graphics Databases Network-Centric Computing Network Security

Teaching

Head TA — Introduction to Artificial Intelligence (CS 331) TA — Introduction to Programming (CS 200, two sections of ~130 students) TA — Computational Problem Solving (CS 100)

Projects

The flagship first, then the AI and machine learning work, then the immersive systems and human-centered research it grew from. More in the downloadable CV.

Founder & Inventor · Jan 2024 – Present

AR-CAD: Augmented-Reality Computer-Aided Design

The augmented-reality CAD system I invented and founded a venture around: design directly in physical space on Meta Quest 3. Validated in a user study and Lab2Market Validate program, with an approved invention disclosure and CAD $25K in commercialization funding through Mitacs.

Coming next: LLM-powered spatial co-design. An AI assistant that designs with you inside your physical space, aware of your scene and context. Stay tuned!!!

Visit AR-CAD Website

AI & Machine Learning

LLM systems, NLP pipelines, computer vision, signal processing, and generative models, each built end to end and evaluated properly.

Jul 2026 · LLM / RAG / Text-to-SQL

Procurement Document Intelligence

An end-to-end LLM system that turns government procurement PDFs into a database you can question in plain English: OCR benchmarking (0.093 character error rate), a heuristic + LLM extraction overlay for bilingual forms, and a RAG text-to-SQL engine (ChromaDB, three LLM providers) with a three-gate SQL safety layer and 156 offline tests.

Live Demo GitHub

May 2025 – Apr 2026 · Presented at Alberta BME Conference

EEG Signal Processing for CAD Cognitive Load

An EEG signal-processing and machine-learning pipeline (MNE-Python) that reconstructs clean cognitive-load spectra from 32-channel recordings taken during live Meta Quest 3 use: FIR filtering, ICA-based artifact removal, marker-based epoching, and Welch bandpower across scalp regions. The thesis designs the next phase as machine-learning-assisted signal reconstruction: Artifact Subspace Reconstruction (statistical, Gaussian-thresholded reconstruction of artifact-contaminated segments), ICLabel (neural-network classification of independent components), and iCanClean adaptive filtering using Meta Quest 3 IMU streams as motion references. A pilot finding, elevated occipital alpha under passthrough, motivated HMD-specific baselines for AR-EEG.

Details

May 2025 - Dec 2025 · Research Square preprint, 2026

Bloom Taxonomy Extension using Machine Learning

A multi-stage NLP pipeline that extended Bloom's cognitive-domain verb inventory from 358 to 1,506 ML-classified engineering verbs using Sentence-BERT embeddings, one-vs-rest classification, and calibration, while preserving the taxonomy's ordinal structure (Spearman rho = 0.90).

Details

Jan 2026 – Apr 2026 · NLP for Design Cognition

Bloom-Coded Think-Aloud Analysis

A Whisper + spaCy NLP pipeline that auto-coded more than 1,800 verb occurrences across 38 think-aloud sessions into Bloom cognitive domains, replacing weeks of manual protocol coding and showing AR-CAD elicited nearly twice the expressed cognitive activity of traditional CAD (Wilcoxon signed-rank with Holm correction).

GitHub

2024 · Deep Learning Time Series

Air Quality Forecasting in Beijing

Benchmarked seven classical regressors against RNN, LSTM, BiLSTM, and GRU networks under blocked cross-validation to prevent temporal leakage; the best RNN reached RMSE 0.536 and R² = 0.947 forecasting pollutant levels and the Air Quality Health Index from multi-station data.

GitHub

Sep 2024 – Dec 2024 · Generative AI

Multi-Style Game Sprite Generation with Diffusion Models

A study comparing diffusion, beta-VAE, and GM-VAE models for pixel-art sprite generation across combined Pokémon and Tiny Hero datasets. Our residual U-Net diffusion model outperformed every VAE baseline on both styles (FID 32.70 vs 47.16), evaluated on 1,000 generated samples per style.

Details GitHub

2024 · Clinical Machine Learning

Machine Learning for Scoliosis Curve Prediction

Benchmarked 10 classification and 10 regression models predicting curve progression and Cobb angle from 211 clinical records, with ensemble methods (Random Forest, AdaBoost) delivering the strongest validated performance.

XR & Human-Centered Research

The immersive systems and user-research foundation behind the AI work: built in Unity, evaluated with real users.

Meditare virtual reality well-being project

Jun 2022 - Jun 2023

Meditare: Virtual Reality for Well-Being

A VR well-being experience built in Unity and SteamVR using immersive environments, 3D interaction, materials, lighting, and environmental effects to support calming and restorative experiences.

Details

Feb 2022 - May 2022

Virtual Reality for Awareness Training

A scenario-based VR awareness and training experience combining research, immersive implementation in Unity, and evaluation of awareness and response readiness.

Details

Sep 2021 - Feb 2022

AssistTH: Telehealth Mobile Application

A user-centered telehealth mobile application connecting patients and physicians, designed through contextual inquiry, interviews, scenarios, prototypes, pilot testing, and usability testing.

Details

Research

Publications & Research Impact

Peer-reviewed research backing the systems and the commercialization around it!

Conference Paper

ASME IDETC/CIE 2025

2025

Lowering Barriers to CAD Adoption: A Comparative Study of AR-CAD and a Traditional CAD Tool

M. Talha, A. Mohiuddin, S. Javed, and A. J. Qureshi. Presented at the 2025 ASME International Design Engineering Technical Conferences & Computers and Information in Engineering Conference (IDETC/CIE).

Peer-reviewed AR/HCI User Study

Read the Paper

Preprint

Research Square, 2026

2026

Extending the Cognitive Domain of Bloom's Taxonomy using Machine Learning

M. Talha, J. Shi, and A. J. Qureshi. Submitted to Discover Education. A Sentence-BERT pipeline extending Bloom's verb inventory from 358 to 1,506 ML-classified engineering verbs while preserving ordinal structure.

NLP Machine Learning Education

Read the Preprint

Let’s connect

Feel free to reach out for software, AR/VR, AI/ML, UX, research, or collaborative opportunities.