Muhammad Uzair — Islamabad, Pakistan
About
I'm a software engineer by training (NUST, 2025) and an AI engineer by trade. My work sits at the intersection of shipped systems and published research — voice AI pipelines, agentic LLMs, and reinforcement-learning policies for wireless networks.
I've spent time benchmarking PPO in cart-pole and pendulum environments, integrating human advisory signals into RL training loops at McMaster, and training multi-agent DRL frameworks for CR-NOMA IoT — work that turned into two IEEE publications.
Day-to-day, I build: voice agents on VAPI + Deepgram, RAG retrieval layers with pgvector, and LLM orchestration services in FastAPI.
Selected work
Domain-specific LLM assistant for banking, built on LLaMA-3.2-3B with LoRA fine-tuning, FAISS retrieval, and real-time document ingestion.
Proximal Policy Optimization agent solving the MuJoCo inverted pendulum, with a hand-shaped reward function and Optuna-driven hyperparameter search.
Privacy-preserving ranking system trained with federated averaging on the ANTIQUE QA dataset, matching centralized baselines without pooling user data.
End-to-end IoT pipeline: ESP32 sensors → Azure IoT Hub → blob storage → ML model → PowerBI dashboards. 75% AQI prediction accuracy on a self-collected dataset.
Peer-reviewed
Saleha Ahmed, Muhammad Uzair, Syed Asad Ullah, et al.
IEEE Internet of Things Journal, 2025
A cooperative multi-agent DRL framework for CR-NOMA IoT, where distributed agents jointly learn spectrum access and power-control policies under partial observability.
Saleha Ahmed, Muhammad Uzair, Syed Asad Ullah, et al.
IEEE International Conference on Communications (ICC), 2025
DRL-driven transmit-power control for energy-harvesting uplink nodes, evaluated against MRC, SC, and EGC diversity-combining schemes under Rayleigh fading.
Timeline
Industry · Islamabad, PK
Nov 2025 — present
Research · Hamilton, ON
Jun 2025 — Aug 2025
Research · Islamabad, PK
Jun 2024 — Sep 2025
Education · Islamabad, PK
Nov 2021 — Jun 2025
Get in touch
Currently open to AI engineering roles and PhD opportunities in reinforcement learning, multi-agent systems, and applied ML.
The fastest way to reach me is email — I usually respond within a day. You can also grab my CV below if you're evaluating me for a role or program.