Resume

Muhammad Huzyafa Khokhar

AI Research Engineer · Multimodal Learning · Autonomous AI Systems · Neuro-Symbolic Reasoning

Last updated 2026 Download Resume

Summary

AI Research Engineer specializing in multimodal learning, autonomous AI systems, and neuro-symbolic reasoning. Developed a sentence-level imagined speech decoding framework from non-invasive EEG and deterministic reasoning systems for regulated industries, while designing autonomous scientific agents that transform complex real-world data into machine-interpretable knowledge.

Research Experience

Research Engineer (Founder) · Excelleve
Neurotechnology & Brain-Computer Interfaces · Oct 2023 – Present
  • Developed a sentence-level imagined speech decoding framework from non-invasive EEG across a 197-sentence vocabulary, achieving 79% Top-5 and 24.5% Top-1 accuracy on the ChiSCO dataset.
  • Designed a dual-encoder contrastive architecture combining a Multi-Scale CNN with a Transformer encoder aligned to frozen BERT representations.
  • Built a topology-driven EEG channel selection framework reducing 122 electrodes to 32 while improving signal quality by 1.92×.
  • Built an EEG-conditioned language generation prototype achieving 0.738 cosine similarity between generated and intended sentences.
Founder & Research Engineer · Glazion
Neuro-Symbolic AI & Trustworthy Machine Learning · Dec 2025 – Present
  • Designed an evolving neuro-symbolic reasoning architecture separating neural perception from symbolic reasoning.
  • Developed an ILP knowledge-discovery framework applied to 1,265 regulatory documents, discovering a compound GMP pattern that eliminated 18 false negatives.
  • Built a hybrid inference pipeline with cryptographically signed HMAC-SHA256 audit trails aligned with 21 CFR Part 11.
  • Achieved 100% decision accuracy and an F1 score of 0.968 across 1,415 real regulatory documents.

Selected Systems

Autonomous Manufacturing Intelligence Agent
  • Autonomous neuro-symbolic research agent transforming manufacturing sensor data into AI-ready datasets, demonstrated on the Peregrine dataset. Technically reviewed by Mark Burhop (Siemens 2022 Inventor of the Year, ASTM F42).
WildNet
  • Multimodal deep-learning architecture for wildlife audio classification, achieving 88% ROC-AUC across 206 bird species under ~1:500 class imbalance.
Universal Sensor Interface for VEX Robotics
  • Reverse-engineered VEX's analog protocol, reducing sensing hardware cost by ~90% while maintaining 100% accuracy across 1,000+ cycles.

Research Expertise

Research Areas: Multimodal Learning, Brain-Computer Interfaces, Neuro-Symbolic AI, Autonomous AI Systems, Representation Learning, Agentic Systems
Machine Learning: PyTorch, Transformers, CNNs, Contrastive Learning, Self-Supervised Learning, Hugging Face
Reasoning & AI Systems: Knowledge Representation, Inductive Logic Programming, Symbolic Reasoning, Hallucination Verification, Tool-Using Agents
Signal Processing: EEG Signal Processing, Persistent Homology (TDA), Phase-Locking Value, Spectral Analysis
Engineering: Python, C++, Docker, AWS, API Design, FAISS, Git

Education

New York University Shanghai
B.S. Computer Science · Minor: Neurolinguistics · Expected May 2028
  • Academic Scholarship. Research Focus: Brain-Computer Interfaces & AI.
International School Lahore
A Levels: Physics, Mathematics, Computer Science, Chemistry, History
  • A2 Markhor Honors Award

Awards

  • Medalist — VEX Asia Open Championship
  • Promising Award — VEX U SJTU World Championship China Selection
  • Top 10 — National Engineering & Robotics Contest
  • Runner-Up — Aitchison Robotics Competition
  • A2 Markhor Honors Award